perf(sync): adaptive quick-fire debounce + scoped watcher sync — save-to-graph well under a second at any scale (#1397)

Two changes to the watcher path (the always-on daemon every agent
session uses), which previously paid a flat 2s debounce plus a full-tree
scan-diff on every save even though the OS events name the exact files:

1. Adaptive debounce: a pending set of ≤2 files fires after a 300ms
   quiet window; ≥3 keeps the full configured window so agent
   multi-file bursts coalesce exactly as before. Re-arming preserves
   trailing-edge semantics; a user-set CODEGRAPH_WATCH_DEBOUNCE_MS
   remains the authoritative upper bound (quick window never exceeds
   it, floor 100ms).

2. Scoped sync: watcher-triggered syncs pass their pending paths, and
   the reconciler stats exactly those — per-path logic identical to the
   full walk (stat pre-filter, hash confirm, the #1240
   removal/resurrection flow) — skipping the O(repo) scan and
   tracked-load. Strict fallbacks keep the full scan-diff as ground
   truth: directory removals (#1285 — the events can't name the
   children), empty pending sets (retry paths), and >500-file storms
   (branch checkouts, which also self-heal anything event coalescing
   dropped). filesChecked counts examined PATHS so a deletion-only
   scoped sync can't mimic the #449 lock-unavailable signature.

Measured (warm in-process, the daemon path): dubbo one-file sync work
512→335ms, Swift compiler (27k files) 884→385ms — save-to-fresh-graph
≈0.6-0.7s end-to-end including the quick debounce, from ~2.5-6s
perceived before. Gates: scoped-vs-full dumps byte-identical on dubbo
AND the Swift compiler; watcher suite 30/30 (3 new: scoped pass-through,
dir-removal fallback, quick-fire timing); sync suite 34/34 (4 new
scoped-parity cases incl. delete-resurrection and the lock signature);
full suite 2,696 ×2 with CODEGRAPH_KERNEL_EXPECT=1.

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

CHANGELOG.md

@@ -9,6 +9,17 @@ and adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 ## [Unreleased]
 
+# ⚡ The Rust engine release
+
+**This release rebuilds CodeGraph's parsing engine as a native Rust kernel and overhauls the resolution pipeline around it — the largest performance upgrade in the project's history, with every graph verified byte-for-byte identical to the previous engine.**
+
+- **Native Rust parsing for 20 languages** — TypeScript, JavaScript (+TSX/JSX), Java, Python, Go, C, C++, Rust, C#, Ruby, PHP, Swift, Kotlin, Scala, Dart, R, Lua, and Luau now parse in a compiled Rust kernel (Metal and CUDA ride the C++ path). Platforms without a prebuilt binary, and individual files with syntax errors, fall back to the previous engine automatically — same graph either way, proven on repositories from small libraries to the Linux kernel.
+- **Adaptive to your machine** — CodeGraph sizes its parse workers, resolver pool, and caches from what the system actually has: real core counts (container/cgroup-aware, not the host's), honest available memory on macOS and Linux, and measured per-project resolution cost. A big workstation gets the full parallel pipeline; a 2-core VPS gets a pipeline tuned to finish reliably instead of running out of memory — the Linux kernel (70k files) indexes to completion on a 2-core, 6GB machine in under 12 minutes — down from 26 at the start of this cycle.
+- **Resolution is dramatically faster across the board** — adaptive parallel resolution, smarter method-candidate lookup, and memoized supertype/conformance walking. The Swift compiler repository (27k files, Swift + C++) went from over 3 minutes to about 100 seconds within this release cycle; Rust, Lua, and Java-family projects all see double-digit improvements.
+- **Sync is now near-instant** — a save reaches the graph in well under a second, even at compiler scale. The always-on watcher fires after a 300ms quiet window for lone saves (bursts of edits still coalesce), and hands the exact changed paths to sync instead of re-scanning the whole tree — measured save-to-fresh-graph work of ~0.3s on a 4,400-file Java project and ~0.4s on the 27,000-file Swift compiler repository, byte-identical to a full reconciliation.
+
+Full details in the entries below.
+
 ### New Features
 
 - Indexing TypeScript, TSX, JavaScript, JSX, Java, Python, Go, C, C++, Rust, C#, Ruby, PHP, Swift, Kotlin, Scala, Dart, R, Lua, and Luau projects is faster: parsing and symbol extraction now run in a native engine when a prebuilt binary is available for your platform (release bundles include one), producing exactly the same graph — verified byte-for-byte against the previous engine on real repositories, from small libraries up to vscode-, dubbo-, django-, git-, protobuf-, tokio-, rust-analyzer-, jellyfin-, rails-, symfony-, swift-nio-, kotlinx.coroutines-, ggplot2-, Kong-, Scala-3-compiler-, and Flutter-scale codebases (Lombok-generated members, C function-pointer tables, and Unreal-Engine-style macro-heavy headers included; CUDA and Metal sources ride the C++ path). The speedup is largest on resource-constrained machines like CI runners. No setup needed: platforms without the native binary, and individual files with syntax errors, automatically use the previous engine, and `CODEGRAPH_KERNEL=0` turns the native path off entirely.
@@ -24,6 +35,7 @@ and adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 - Indexing very large projects on multi-core machines got faster again: the parallel-resolution workers now periodically refresh their read-only database connections, which lets database housekeeping advance instead of silently building up a backlog behind long-lived readers — a backlog that was taxing the indexer's own writes. Graphs remain byte-for-byte identical; the win is largest at Linux-kernel scale on many-core machines.
 - Indexing on macOS now uses the machine's real memory headroom when sizing its parallel-resolution workers. macOS deliberately keeps RAM filled with reclaimable cache, so the previous free-memory reading came back tiny (~1GB on an otherwise idle machine) and silently halved the worker pool — a medium Java project's fresh index ran about 15–20% slower than the hardware allowed. Graphs remain byte-for-byte identical; the same fix also lets a memory-driven analysis cache engage fully on macOS for large C codebases.
 - Fresh indexing got a sizeable across-the-board speedup: during the initial build, the database's secondary lookup indexes are set aside and rebuilt once after parsing instead of being maintained row by row — the same proven trick the later linking phase already used, now applied to the whole parse lane — and the reference-resolution loop likewise stops maintaining lookup indexes it never reads, rebuilding them at the end when almost nothing is left in the table. A medium Java project's parse phase runs about 58% faster and its full fresh index about 19% faster end-to-end; a Linux-kernel-scale index that took ~15 minutes on an 8-core machine now completes in about 11, with the resolution phase alone dropping by a third. Graphs remain byte-for-byte identical, and incremental syncs are unaffected.
+- Saving a file now updates the graph almost immediately: the file watcher fires after a 300ms quiet window for one or two changed files (bursts still coalesce under the full debounce, and `CODEGRAPH_WATCH_DEBOUNCE_MS` remains the upper bound), and watcher-triggered syncs reconcile exactly the changed paths instead of stat-walking the entire repository. Directory deletions and event storms still run the full scan-diff, so nothing the events can't describe is ever missed. Measured: save-to-fresh-graph sync work drops to ~0.3s on a 4,400-file project and ~0.4s on a 27,000-file one, with resulting graphs byte-for-byte identical to a full reconciliation.
 - Indexing Swift and other protocol/interface-heavy codebases got dramatically faster: the conformance walk that checks whether a method lives on a receiver's supertypes (protocols, base classes, extensions) now remembers its answers for the duration of each resolution batch instead of re-querying the graph for every call site — on the Swift compiler repository (27k files) that walk ran nearly a million times per index. A fresh index of that repo drops from about 185 seconds to under 100, with the graph byte-for-byte identical. Method-candidate lookup also gained a per-name owner index, so overload-heavy names (`init` in Swift, `execute` in Java) no longer pay a full candidate scan per receiver type.
 - Resolving method calls through local variables (`recv.method()`, Lua's `recv:method()`, R's `recv$method()`) got much cheaper on repos where the same receiver is called over and over: the declaration scan that types the receiver now remembers what it has already scanned per scope instead of re-reading the same source lines for every call site, and the regex patterns it scans with are compiled once per receiver instead of per call. Kong's fresh index drops another 8% on top of the require-resolution fix (23% cumulative), with graphs byte-for-byte identical everywhere — including Java projects, where this same scan successfully types tens of thousands of receivers.
 - Indexing Lua and Luau projects got a sizeable speedup: resolving each `require(...)` no longer rescans the project's entire file list four times — a per-project filename index answers the same lookup instantly, cutting per-require resolution from about a millisecond to microseconds. A fresh index of Kong (1,870 Lua files) runs about 16% faster end-to-end, with the graph byte-for-byte identical. The same housekeeping also closes a latent staleness edge where COBOL copybook lookups could keep serving a cached file list after files changed.