fix(mcp): route FalkorDB UUID operations to tenant graphs

[noamsiegel]

Summary

• Resolve UUID-keyed MCP reads and deletes against the FalkorDB graph that owns the requested data.
• Reject UUIDs found in multiple tenant graphs rather than reading or deleting an arbitrary match.
• Batch provenance resolution with one probe per graph and avoid FalkorDriver.clone() index-build side effects during reads.
• Serialize queued cross-group writes because Graphiti.add_episode() mutates the shared driver binding.

Problem

FalkorDB stores each group_id as a physical graph. The remaining UUID-keyed MCP handlers bypass group-aware routing and query whichever graph the shared driver currently targets. Concurrent per-group queue workers can also race while Graphiti.add_episode() rebinds that shared driver. Both cases can report missing data or operate on the wrong tenant graph.

get_episodes is intentionally not changed here because current main already routes EpisodicNode.get_by_group_ids through the merged FalkorDB operations fix.

Fixes #1651.

Verification

• python -m pytest -q tests/test_falkor_routing.py tests/test_core_parity.py - 34 passed.
• python -m ruff format --check ... and python -m ruff check ... - clean.
• python -m pyright --pythonpath <venv-python> src/utils/falkor_routing.py src/graphiti_mcp_server.py src/services/queue_service.py - 0 errors.
• Mutation checks fail when UUID matching, duplicate detection, non-initializing driver scoping, or cross-group write serialization is removed.
• Live FalkorDB reproduction confirmed cross-graph edge lookup, cross-graph episode deletion, and distinct-group concurrent writes on the patched MCP service.

[zep-cla-assistant]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[noamsiegel]

I have read the CLA Document and I hereby sign the CLA behalf on myself, e-mail: noam@noamsiegel.com

[noamsiegel]

I have read the CLA Document and I hereby sign the CLA