fix: stabilize HTTP MCP indexing

[linhdmn]

Summary

• Stop HTTP MCP tool calls from doing surprise per-request auto-index work and serialize write-heavy tools.
• Make mcp_index skip expensive global call-edge resolution by default via resolve_calls: false.
• Repair legacy Cozo graph schemas where code_elements/relationships were missing the env column even after migration 006 was recorded.
• Make mcp_status lightweight by default and add include_counts for full counts.
• Fix project path routing, nested-runtime health checks, nested worktree indexing, and malformed graph queries found during the investigation.

Validation

• cargo fmt --check
• cargo check
• cargo build
• cargo test test_init_db_repairs_legacy_code_elements_after_recorded_migration -- --nocapture
• cargo test test_mcp_status -- --nocapture
• cargo test test_mcp_index -- --nocapture
• cargo test mcp_server -- --nocapture
• cargo test test_find_files -- --nocapture
• git diff --check
• Manual HTTP MCP validation on /Users/linh.doan/work/be for mcp_status, search_code, find_function, and mcp_index.

Notes

• The BE workspace DB schema was repaired during local validation.
• A fixed local server is currently running on port 9699 from tmux session leankg-mcp-9699-fixed for Cursor validation.