feat: MCP client identity + memory diagnostics (#37)

[justrach]

Summary

MCP client identity (#37)

• Extracts clientInfo.name from MCP initialize params
• Stores in Session.client_name for future agent identity use
• Foundation for multi-client conflict detection

Memory diagnostics in codedb_status

Inspired by fff.nvim's health reporting. codedb_status now returns:

codedb status:
  seq: 84
  files: 84
  outlines: 84
  contents_cached: 84
  trigram_index: mmap (84 files)
  index_memory: 6586KB

Shows: outline count, cached content count, trigram index type (heap/mmap/mmap+overlay), indexed file count, and approximate index memory in KB.

Other

• Fixed handleSnapshot that was accidentally deleted
• Updated serverInfo version to 0.2.54

Test plan

• All tests pass (zero leaks)
• Live MCP test confirms new status fields
• Pre-push benchmarks pass

Closes #37

[github-actions]

Benchmark Regression Report

Threshold: 10.00%

Tool	Base (ns)	Head (ns)	Delta	Status
codedb_bundle	642255	656198	+2.17%	OK
codedb_changes	107675	109861	+2.03%	OK
codedb_deps	30324	31162	+2.76%	OK
codedb_edit	20494	41515	+102.57%	FAIL
codedb_find	141199	140868	-0.23%	OK
codedb_hot	160665	170689	+6.24%	OK
codedb_outline	413014	416085	+0.74%	OK
codedb_read	168530	174067	+3.29%	OK
codedb_search	261494	260461	-0.40%	OK
codedb_snapshot	3922377	3899865	-0.57%	OK
codedb_status	256794	253491	-1.29%	OK
codedb_symbol	63620	65148	+2.40%	OK
codedb_tree	99290	104861	+5.61%	OK
codedb_word	93353	91673	-1.80%	OK

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f7488f8b25

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]

Hold explorer read lock while sizing indexes

handleStatus calls telemetry_mod.approxIndexSizeBytes(explorer) before acquiring explorer.mu.lockShared(). That helper iterates mutable index/hash-map structures, so a concurrent writer (initial/background scan or file watcher updates that take the write lock) can mutate those containers mid-iteration and trigger undefined behavior or crashes during codedb_status. The existing recordCodebaseStats path already wraps this same helper under explorer.mu.lockShared(), so this unlocked call is a regression in thread-safety.

Useful? React with 👍 / 👎.