feat(mcp): codegraph_explore as the sole primary tool + store coverage + overload disambiguation (#647)

## Summary

Completes the explore-overhaul arc: `codegraph_explore` becomes the single primary tool an agent reaches for, and its coverage + output shape are tuned so flow/architecture questions resolve with near-zero Read/Grep.

### What changed
- **explore is the sole primary tool** — removed `codegraph_context` (the fuzzy-input Read-trigger) and `codegraph_trace` (under-picked by agents); explore already surfaces the call flow among the symbols you name. A plain natural-language question now works as the query.
- **Store/handler coverage** — functions defined inside object literals (Zustand `create((set, get) => ({ … }))`, Redux/Pinia/MobX, exported handler/route maps) are indexed as real symbols, including calls through `useStore.getState().fn()` and destructured `const { fn } = useStore.getState()`. A general AST rule, not a per-lib hack.
- **Overload disambiguation** — explore leads with the *right* definition when a method name is overloaded across types (a PascalCase type token in the query biases to that type's own def); `codegraph_node` returns *every* overload's body in one call, with an optional `file`/`line` selector to pin one.
- **Method-atomic render** — explore never returns half a method; at the size budget it drops whole methods/files (and lists what it dropped) instead of truncating a body mid-method.
- **Native-read-shaped output** — per-call output is capped to ~24K with a 25K hard ceiling and concentrated into ~150–250-line flow windows, mirroring how the agent natively reads; repo size scales the *call* budget, not the per-call size (a larger response just gets externalized to a file the host Reads back).
- **Blast radius** folded into explore (dependents + covering tests, locations only).

### Benchmark (refreshed on this build)
Re-validated the 7-repo A/B on 2026-06-02 (Opus 4.8, effort=high, median of 4). WITH arm re-measured on this build, WITHOUT reused:

**~16% cheaper · 47% fewer tokens · 22% faster · 58% fewer tool calls** — 0 file reads on 6 of 7 repos (Gin ~1).

The arc trades larger, cache-heavy explore responses for guaranteed near-zero reads, so cost/token margins soften vs the prior build (Excalidraw and Tokio land at cost break-even) while time and tool-calls stay clear wins everywhere — consistent with the project's stated optimization target (latency + tool-calls, not token cost).

### Validation
- Full suite green: **1112 passed, 2 skipped**.
- 28/28 plain WITH runs across the 7 README repos completed clean; reads median 0 on 6/7.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

CHANGELOG.md

@@ -9,6 +9,18 @@ and adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 ## [Unreleased]
 
+### New Features
+
+- `codegraph_explore` is now the primary tool, and one call is usually all an agent needs: it returns the verbatim source of the symbols relevant to your question (a plain question works as the query — you no longer need exact symbol names), grouped by file and Read-equivalent, so the agent answers without falling back to read/grep. The narrower `codegraph_context` and `codegraph_trace` tools were removed in favor of it — explore already surfaces the call flow among the symbols you name (the job trace did), so there's one obvious tool to reach for instead of three.
+- `codegraph_explore` now includes a compact "Blast radius" for the symbols you're looking at — who depends on each (just the locations, not their source) and which test files cover it — so before editing, the agent can see what else to update and which tests to run, without a separate impact lookup. Symbols nothing depends on are skipped, so it stays short.
+- Functions defined inside a store or handler object — the actions in a Zustand `create((set, get) => ({ … }))` store, and the same shape in Redux, Pinia, MobX, or any exported handler/route map — are now indexed as real symbols. Previously they existed only as object properties, so looking one up by name or asking who calls it returned "not found" and the agent had to read the whole store file to follow the flow; now `codegraph_node`, `codegraph_callers`, and `codegraph_explore` resolve them directly — including calls made through `useStore.getState().fetchUser()` or a destructured `const { fetchUser } = useStore.getState()`.
+- `codegraph_explore` now surfaces the *right* definition when a method name is overloaded across types. Asking about, say, `DataRequest`'s `task` and `validate` used to return a same-named method from an unrelated file (or an abstract base stub) and bury the one you meant; explore now recognizes the type you named in the query and leads with that type's own overloads, in full.
+
+### Fixes
+
+- Search ranking no longer lets a common word in your request hijack the results: asking about, say, a "flat object" screen used to surface an unrelated constant that merely happened to be named the same, because the exact-name match outweighed everything else. Ranking now weighs how well each result is corroborated by the rest of your request, so the symbols you actually meant come first (this improves `codegraph_explore`'s results).
+- `codegraph_node` now returns *every* definition when a name is ambiguous — an overloaded method, or the same method name on different types — instead of returning one (sometimes the wrong one) with a note listing the rest. Asking for such a symbol now hands back all of the matching definitions with their source in a single call, so the agent stops having to read the file by hand to find the specific overload it wanted (common in Swift, Go, Java, and C#). For a heavily-overloaded name (a `poll`/`validate` with dozens of definitions), pass `file` (and/or `line`) — e.g. the `file:line` shown in a trail — to get that exact definition's body. Large overload sets show the most relevant ones in full and list the remainder by location.
+- `codegraph_explore` never returns half a method anymore: when output runs up against its size budget it drops whole methods or whole files (and lists what it dropped, so you can ask for them in another call) instead of cutting off a method body partway. A truncated method was the one case that still sent the agent to read the file for the rest — so the source explore returns is now always complete and usable as-is.
 
 ## [0.9.8] - 2026-06-01