fix: _filter_under_community_level silently drops entities without community assignments

[hanhan761]

Fixes #2348

Problem

_filter_under_community_level in indexer_adapters.py filters entities using df[df.level <= community_level]. Entities without community assignments have NaN in the level column after the left join with community membership data. Since NaN <= x evaluates to False in pandas/NumPy, all unassigned entities are silently dropped with no warning.

This causes CLI query commands (graphrag query --method local/global/drift) to operate on a drastically reduced entity set. Small, sparse, or domain-specific datasets are especially vulnerable since isolated nodes are routinely excluded by Leiden community detection.

Fix

Changed the filter to also preserve rows where level is NaN:

df[(df.level <= community_level) | df.level.isna()]

This is semantically correct: "not in any community" is not the same as "in a community above the requested level." Entities without community assignments should pass through the filter.

Test

Added tests/unit/query/test_indexer_adapters.py with 5 test cases covering:

• NaN levels preserved alongside valid levels
• Entities above the threshold correctly filtered out
• All entities below threshold preserved
• All NaN (no community assignments) preserved
• Mixed NaN and above-threshold levels