fix(falkordb): route namespace/ops get_by_group_ids reads to the group's graph

[spencer2211]

Summary

#1670 fixed single-group_id routing for the decorated path (@handle_multiple_group_ids, which clones self.clients.driver and passes it down as driver=). The namespace/ops read path is a separate implementation and is still unrouted on main.

graphiti.nodes.episode.get_by_group_ids(...) delegates to the FalkorDB *Operations classes (FalkorEpisodeNodeOperations.get_by_group_ids etc.). Those methods receive the base driver as executor — pointed at default_db — and run the query directly, so they never go through the decorator's clone. On FalkorDB, where each group_id is its own graph, a single-group read therefore queries the empty default graph and returns 0 rows.

This is the path reported in #1651 (get_episodes / delete_episode querying the wrong graph from the MCP server). There are currently no clone call sites anywhere under graphiti_core/driver/falkordb/operations/.

Related history: #1325 is the original report; #1326 covered both the classmethod path and this ops path, and was closed in favour of #1670, which covered only the former. This PR carries the remaining half forward against current main.

Change

Apply single-group executor.clone(database=gid) routing plus multi-group fan-out to all nine FalkorDB operations get_by_group_ids methods:

• nodes: episode, entity, community, saga
• edges: episodic, entity, community, has_episode, next_episode

The clone is guarded by isinstance(executor, GraphDriver) so the QueryExecutor parameter narrows to a clone-capable driver (the QueryExecutor protocol does not declare clone/provider), keeping pyright clean.

This is complementary to #1670, not overlapping: the decorator routes at the search/retrieve entry point via driver=, this routes at the ops layer via executor=.

Test plan

Adds tests/driver/test_falkordb_ops_routing.py — a package-free regression using a spec=GraphDriver mock, so it needs no live FalkorDB and runs in CI. Covers single-group routing, multi-group fan-out, and the empty-group_ids pass-through.

• uv run pytest tests/driver/test_falkordb_ops_routing.py -q → 5 passed
• uv run pytest tests/driver/ tests/test_handle_multiple_group_ids.py -q → 34 passed, 1 skipped (includes fix(falkordb): route single group_id search to the matching graph #1670's decorator tests — no interaction regression)
• make lint → ruff clean, pyright 0 errors

Happy to consolidate the repeated fan-out into a shared helper rather than one block per ops class if you'd prefer that shape.

[zep-cla-assistant]

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

[spencer2211]

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

[spencer2211]

@Naseem77 this is the namespace-path gap you flagged on #1326 — carried forward onto post-#1670 main as a standalone change, since #1670 covered the decorated path only. Same nine ops methods and the same package-free regression test you reviewed there.

cc @prasmussen15 for context: #1326 was closed in favour of #1670, which fixed @handle_multiple_group_ids but not the *Operations.get_by_group_ids read path (#1651).

[spencer2211]

I have read the CLA Document and I hereby sign the CLA behalf on myself, e-mail: spencer.p.fuller@gmail.com

[spencer2211]

recheck