Pre-release: harden Gin METHOD path route-query sufficiency

[jununfly]

This was generated by AI during triage.

Parent

• References Rust-hybrid pre-release: TS/JS and Go Agent Sufficiency refresh #279: TS/JS and Go Agent Sufficiency refresh found mixed Go/Gin results.
• References Rust-hybrid pre-release: final closeout decision #280: pre-release closeout decision accepted the Go/Gin caveat as non-blocking.

What to build

Make zcodegraph_explore more sufficient for Go/Gin route lookup questions that mention an HTTP method and path, such as POST /upload. When the query contains a METHOD path shape, Explore should prioritize matching existing route nodes and surface the route-to-handler relationship and relevant handler source so the agent has enough evidence to answer without reading the small route file.

This is a bounded pre-release hardening attempt. It should prefer using the existing graph and Explore planning path over changing the Rust Go extractor.

Non-goals:

• Do not change the Rust Go extractor unless the deterministic tool-level test proves the existing graph lacks the needed route/handler data.
• Do not add Go module/package resolution.
• Do not expand Gin middleware-chain semantics.
• Do not change MCP tool names or protocol.
• Do not run a full benchmark suite.

Acceptance criteria

• Add deterministic tool-level coverage showing zcodegraph_explore on a Gin POST /upload style query returns the matching route node and handler evidence.
• Preserve existing broad Gin route overview behavior.
• Run the targeted real Gin Agent Sufficiency A/B prompt from Rust-hybrid pre-release: TS/JS and Go Agent Sufficiency refresh #279: How does a request reach the upload handler for POST /upload?
• If A/B improves, update README and the pre-release closeout decision with the improved Go/Gin evidence.
• If deterministic coverage passes but A/B still reads main.go, record the conclusion as: mechanism available, agent still falls back; no release blocker.
• If deterministic coverage cannot pass without a larger extractor/resolver change, record a no-go decision and do not expand scope in this issue.
• Do not add a second optimization candidate in this issue.

Blocked by

None - can start immediately.

[jununfly]

Completed in fc55171. Added bounded Gin METHOD/path route-query hardening for zcodegraph_explore without changing the Rust Go extractor, Go package resolution, MCP protocol, or middleware semantics. Deterministic tool-level test passes: zcodegraph_explore on a Gin POST /upload fixture now returns the route, handler, registration line, and Route matches evidence. Real targeted A/B improved from the #279 baseline: WITH ZCodeGraph is now 20s / 1 tool / 0 Read / 0 Bash; WITHOUT is 62s / 12 tools / 6 Read / 5 Bash. Evidence: docs/benchmarks/2026-06-19-rust-hybrid-issue-281-gin-route-query-sufficiency.md. Validation: npm run build; targeted explore/Gin tests; full npm test passed (1835 passed, 15 skipped).