fix(communities): scope remove_communities to the group_ids being rebuilt

[Saltasm]

Problem

Fixes #1657.

build_communities(group_ids=[...]) documents scoped behavior — "Create communities only for the listed group_ids" — but its clear step deletes every Community node in the database:

# graphiti.py — build_communities
await remove_communities(driver)   # no group filter

MATCH (c:Community)
DETACH DELETE c

The rebuild then recreates communities only for the selected groups, so a scoped rebuild silently destroys all other groups' communities. Observed live on a 5-group graph: rebuilding 2 stale groups deleted the other 3 groups' 71 communities with no error or log line. Any multi-tenant deployment refreshing communities per tenant hits this on every call, and recovery costs a full LLM re-summarization across all groups.

Fix

Thread group_ids through the clear step:

• community_operations.remove_communities(driver, group_ids=None) — scoped DETACH DELETE with WHERE c.group_id IN $group_ids when group_ids is provided; the existing full wipe when it is not (blank/None keeps the documented whole-graph behavior).
• Same optional parameter on GraphOperationsInterface.remove_communities and all four driver implementations (neo4j, falkordb, kuzu, neptune) — each mirrors the group_ids is None branching already used by sibling operations in the same files.
• build_communities passes its group_ids through.

Tests

tests/utils/maintenance/test_remove_communities.py (mocked driver, no DB):

• unscoped call still deletes all (unchanged behavior)
• scoped call emits the WHERE c.group_id IN $group_ids form with the right params
• group_ids=[] behaves like None (no scoping requested, matching build_communities' blank semantics)
• the graph_operations_interface fast path receives group_ids

🤖 Generated with Claude Code