fix: correctly flatten update_community results in add_episode

[octo-patch]

Fixes #1396

Problem

add_episode crashes with ValueError: too many / not enough values to unpack when the number of extracted entity nodes is anything other than exactly 2.

The root cause: semaphore_gather returns a list of N tuples (one per coroutine), where each tuple is (list[CommunityNode], list[CommunityEdge]). The previous code did:

communities, community_edges = await semaphore_gather(...)

This Python unpacking treats the returned list as if it has exactly 2 items, so it only worked by accident when N == 2. For N == 1 it raises ValueError: not enough values to unpack; for N >= 3 it raises ValueError: too many values to unpack.

Solution

Replace the tuple unpack with an explicit accumulation loop:

per_node_results = await semaphore_gather(...)
for node_communities, node_community_edges in per_node_results:
    communities.extend(node_communities)
    community_edges.extend(node_community_edges)

This correctly aggregates results across all nodes regardless of how many entity nodes were extracted from the episode. Also adds explicit type annotations to communities and community_edges since they were previously inferred as list[Any].

Testing

The fix is a straightforward logic correction with no behaviour change for callers — the returned communities and community_edges lists now have the correct types and values for all N. No external dependencies are required to verify the logic.

[danielchalef]

Thanks for this — correct fix for the update_community results not being flattened from semaphore_gather in add_episode (graphiti.py:1185). Consolidating on #1486 as the canonical fix (same fix + a regression test, rebased on main). Closing as superseded by #1486; credit to you for surfacing it too.