feat: auto-index on first tool call when project is not yet indexed

[maplenk]

Summary

When any query tool is called and the project isn't indexed yet, the server now auto-indexes it inline instead of returning "not indexed". This eliminates the manual index_repository step for first-time use — users can just start asking questions.

The existing background auto-index at startup is preserved; this adds synchronous fallback when a tool call arrives before background indexing completes (or when it wasn't triggered).

Changes

• try_auto_index() — consolidates resolve_store + verify_project_indexed into a single function that auto-indexes on cache miss. Respects auto_index config flag and auto_index_limit file count guard. Uses pipeline lock to avoid concurrent indexing. Waits for any in-progress background auto-index thread before falling back to inline indexing.
• session_project_alt — handles macOS /tmp → /private/tmp symlink variants so tool calls with either project name resolve correctly
• cbm_mcp_server_set_session_root() — public API for external session injection (derives both primary and alt project names)
• 6 handlers modified to use try_auto_index(): get_graph_schema, search_graph, query_graph, get_architecture, trace_call_path, get_code_snippet

Guards

• Config-gated: requires auto_index = true (default: false)
• File limit: rejects repos exceeding auto_index_limit (default: 50,000 files)
• Shell-safe path validation before git ls-files
• Pipeline lock prevents concurrent indexing

Blocking behavior

On a cache miss, the first query can block while inline indexing runs; this path is gated by auto_index=true and auto_index_limit.

Test plan

• Integration test: creates temp project, does NOT pre-index, calls search_graph → verifies auto-index triggers and returns results
• All 2587 tests pass (existing + new)
• Build clean on macOS arm64

[maplenk]

These changes are based on other open PRs. Majorly #151

[DeusData]

@maplenk — auto-indexing on first tool call is an interesting idea, but this is a UX/behavior change that needs discussion first. Please open an issue. See CONTRIBUTING.md.

The current explicit index_repository call is intentional — it gives the user control over when indexing happens and avoids surprise latency on the first query.