test(ebook): cover protected phrases and migrate suite to async

Add test_protected_phrases.py covering phrase-matching behavior in the
RAG engine, and update the existing ebook_search tests to use the async
SQLAlchemy engine/session (create_async_engine, AsyncSession) and async
HTTP paths.

tests/ebook_search/test_rag_pipeline.py

@@ -2,10 +2,11 @@
 
 from __future__ import annotations
 
+import asyncio
 from threading import Event
 from typing import TYPE_CHECKING
 
-from sqlalchemy import create_engine
+from sqlalchemy.ext.asyncio import create_async_engine
 
 from python.ebook_search.config import EbookSearchConfig, RerankConfig
 from python.ebook_search.search import SearchResult, search_ebooks
@@ -14,18 +15,18 @@ if TYPE_CHECKING:
     from pytest_mock import MockerFixture
 
 
-def test_search_ebooks_runs_vector_and_bm25_in_parallel(mocker: MockerFixture) -> None:
-    engine = create_engine("sqlite+pysqlite:///:memory:", future=True)
+async def test_search_ebooks_runs_vector_and_bm25_in_parallel(mocker: MockerFixture) -> None:
+    engine = create_async_engine("sqlite+aiosqlite:///:memory:")
     vector_started = Event()
     bm25_started = Event()
     received_engines: list[object] = []
 
-    def fake_vector_candidates(received_engine, query, _config):
+    async def fake_vector_candidates(received_engine, _client, query, _config):
         """Return vector candidates after confirming BM25 has started."""
         received_engines.append(received_engine)
         assert query == "what is parallel"
         vector_started.set()
-        assert bm25_started.wait(timeout=2)
+        assert await asyncio.to_thread(bm25_started.wait, 2)
         return [SearchResult(chunk_id=1, text="vector", source_title="Vector", vector_score=0.9)]
 
     def fake_bm25_candidates(query, _config):
@@ -39,7 +40,7 @@ def test_search_ebooks_runs_vector_and_bm25_in_parallel(mocker: MockerFixture) -
     mocker.patch("python.ebook_search.search.bm25_candidates", side_effect=fake_bm25_candidates)
     config = EbookSearchConfig(rerank=RerankConfig(enabled=False))
 
-    response = search_ebooks(engine, "what is parallel", config)
+    response = await search_ebooks(engine, mocker.Mock(), "what is parallel", config)
 
     timings = {step.name: step for step in response.timings}
     assert [result.chunk_id for result in response.results] == [1, 2]
@@ -47,3 +48,26 @@ def test_search_ebooks_runs_vector_and_bm25_in_parallel(mocker: MockerFixture) -
     assert timings["BM25 search"].counts_toward_total is False
     assert timings["Hybrid retrieval"].counts_toward_total is True
     assert received_engines == [engine]
+
+
+async def test_search_ebooks_skips_phrase_matching_when_disabled(mocker: MockerFixture) -> None:
+    """Phrase matching can be disabled for one search request."""
+    engine = create_async_engine("sqlite+aiosqlite:///:memory:")
+    mocker.patch(
+        "python.ebook_search.search.vector_candidates",
+        return_value=[SearchResult(chunk_id=1, text="vector", source_title="Vector", vector_score=0.9)],
+    )
+    mocker.patch("python.ebook_search.search.bm25_candidates", return_value=[])
+    detect_mock = mocker.patch("python.ebook_search.search.query_phrase_matches")
+    boost_mock = mocker.patch("python.ebook_search.search.apply_phrase_mention_boosts")
+    config = EbookSearchConfig(rerank=RerankConfig(enabled=False))
+
+    response = await search_ebooks(engine, mocker.Mock(), "what is parallel", config, phrase_matching=False)
+
+    timing_names = {step.name for step in response.timings}
+    assert [result.chunk_id for result in response.results] == [1]
+    assert response.phrase_matches == ()
+    assert "Protected phrase detection skipped" in timing_names
+    assert "Phrase mention boost skipped" in timing_names
+    detect_mock.assert_not_called()
+    boost_mock.assert_not_called()