Cache snapshot responses by store sequence

[justrach]

Summary

• cache codedb_snapshot responses per default/project context using Store.currentSeq() as the freshness key
• invalidate automatically when edits/indexing advance the store sequence
• keep a 16 MB cap so very large snapshots do not stay pinned in memory
• add regression coverage that repeated snapshots are reused and a changed store seq rebuilds the output

Local benchmark

Compared this branch against current origin/release/0.2.579 after #325:

• codedb_snapshot: 602450 ns -> 68050 ns (-88.70%, -534400 ns)

This stacks on top of #325's JSON-generation improvement. After this change, repeated snapshot calls mostly pay for copying the cached response into the MCP output buffer.

Validation

• zig build test
• zig build
• python3 scripts/run-bench-json.py /tmp/codedb-cache-head.json
• base comparison via /tmp/codedb-release-cache-base at origin/release/0.2.579

[github-actions]

Benchmark Regression Report

Thresholds: 10.00% and 50,000 ns absolute delta

NOISE means the percentage threshold was exceeded, but the absolute delta was too small to fail CI.

Tool	Base (ns)	Head (ns)	Delta	Abs Delta (ns)	Status
codedb_bundle	480247	481286	+0.22%	+1039	OK
codedb_changes	53588	54485	+1.67%	+897	OK
codedb_deps	9279	8965	-3.38%	-314	OK
codedb_edit	5978	6397	+7.01%	+419	OK
codedb_find	62833	62174	-1.05%	-659	OK
codedb_hot	93997	96909	+3.10%	+2912	OK
codedb_outline	236476	240485	+1.70%	+4009	OK
codedb_read	85215	85577	+0.42%	+362	OK
codedb_search	175316	174314	-0.57%	-1002	OK
codedb_snapshot	1339408	256474	-80.85%	-1082934	OK
codedb_status	216829	220450	+1.67%	+3621	OK
codedb_symbol	60583	59715	-1.43%	-868	OK
codedb_tree	76959	79337	+3.09%	+2378	OK
codedb_word	68132	68504	+0.55%	+372	OK

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d99041bb0b

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Avoid caching snapshot before explorer catches up

handleSnapshot reads store.currentSeq() and uses it as the cache key before building the snapshot; that assumes explorer state is already synchronized with that seq. In the watcher path, incrementalDiff calls store.recordSnapshot(...) first and only then runs indexFileContent(...) (src/watcher.zig around lines 1001-1007), so a concurrent codedb_snapshot can cache pre-update explorer data under the new seq. Once this happens, later snapshot calls with that same seq will keep returning stale output until another write advances the seq, which is a user-visible correctness regression.

Useful? React with 👍 / 👎.