fix: recover stale docker sync owners

[eunaverse]

Summary

• fix Docker same-PID stale sync owner recovery so previous container generations do not keep source syncs permanently blocked
• add runtime/startup regressions for same-PID preserve and stale-recover cases in MetadataStore
• document the source-sync guard behavior and record the bounded local metadata cleanup evidence

Review notes

• Claude Desktop should be the only process/container starting this MCP server against the shared contextwiki_data volume
• the bounded local metadata repair cleared leftover running sync rows without touching document/chunk content or Chroma data

Verification

• uv run --locked pytest -q tests/storage/test_metadata_store.py -k "same_pid_previous_owner or stale_previous_owner_even_if_pid_looks_alive or live_previous_owner or previous_owner or begin_sync_job_preserves_same_pid_previous_owner or begin_sync_job_recovers_stale_same_pid_previous_owner"
• uv run --locked pytest -q tests/storage/test_metadata_store.py
• python -m compileall api core environments fetching indexing search storage main.py
• ./scripts/verify_functional_e2e.sh

관련 이슈 없음