fix(recall): decouple temporal seed threshold

[Sanderhoff-alt]

Summary

Fixes #2594.

This keeps min_scores.semantic scoped to the semantic retrieval arm. Temporal retrieval uses embeddings only to choose time-window entry points, so reusing the request-level semantic floor there made temporal recall unexpectedly narrower when callers only wanted to prune weak semantic matches.

The change restores the temporal entry-point default threshold of 0.1 while leaving semantic and BM25 request floors unchanged.

Tests

.venv/bin/python -m pytest hindsight-api-slim/tests/test_temporal_recall_selection.py

[benfrank241]

lgtm — right call. min_scores.semantic should scope to the semantic arm, not silently tighten temporal seeds, and 0.1 was already the default so non-min_scores callers see no change. also lines it up with graph, which keeps its own fixed 0.3 seed.

one ask before merge: add a test that pins the decoupling. test_temporal_recall_selection.py doesn't currently touch min_semantic, so nothing stops this getting re-coupled later. a small one — recall with min_scores.semantic=0.5 and assert the temporal call still gets semantic_threshold=0.1 — is enough add that and i'll merge.

[Sanderhoff-alt]

@benfrank241 thanks, added the regression test you suggested. It exercises recall with min_semantic=0.5 and verifies the temporal path still gets semantic_threshold=0.1, so the semantic min score stays scoped to the semantic arm and won’t tighten temporal seeds again.