feat(recall): add opt-in BM25 query term cap

[r266-tech]

Summary

Fixes #2528 by adding an opt-in cap for PostgreSQL native BM25 query terms.

Native to_tsquery currently OR-joins every normalized query token. For very long recall/consolidation queries, that can produce a huge query that matches too much of a large bank. This adds HINDSIGHT_API_BM25_MAX_QUERY_TERMS so deployments can bound the native tsquery term list.

Compatibility notes:

• The default is 0, which preserves the existing uncapped behavior.
• Positive values cap only PostgreSQL native BM25 term preparation.
• VectorChord, pg_textsearch, PGroonga, and pg_search continue to receive the raw query text.
• Oracle accepts the expanded dialect argument but keeps its existing behavior.

Tests

uv run --python 3.12 --project hindsight-api-slim --group dev ruff format --check hindsight-api-slim/hindsight_api/engine/search/retrieval.py hindsight-api-slim/hindsight_api/config.py hindsight-api-slim/hindsight_api/engine/sql/base.py hindsight-api-slim/hindsight_api/engine/sql/postgresql.py hindsight-api-slim/hindsight_api/engine/sql/oracle.py hindsight-api-slim/tests/test_multilingual_bm25.py hindsight-api-slim/tests/test_recall_config.py
uv run --python 3.12 --project hindsight-api-slim --group dev ruff check hindsight-api-slim/hindsight_api/engine/search/retrieval.py hindsight-api-slim/hindsight_api/config.py hindsight-api-slim/hindsight_api/engine/sql/base.py hindsight-api-slim/hindsight_api/engine/sql/postgresql.py hindsight-api-slim/hindsight_api/engine/sql/oracle.py hindsight-api-slim/tests/test_multilingual_bm25.py hindsight-api-slim/tests/test_recall_config.py
uv run --python 3.12 --project hindsight-api-slim --group dev pytest hindsight-api-slim/tests/test_multilingual_bm25.py hindsight-api-slim/tests/test_recall_config.py -q

Codex adversarial review: approved after reworking the cap to be opt-in and keeping legacy config/dialect behavior compatible.