refactor(ebook-search): simplify search and phrase matching

tests/ebook_search/test_http.py

@@ -10,6 +10,7 @@ import pytest
 from python.ebook_search.answer import answer_query
 from python.ebook_search.config import EbookSearchConfig, RerankConfig
 from python.ebook_search.embeddings import embed_texts
+from python.ebook_search.llm_interface import check_chat_endpoint, check_embedding_endpoint
 from python.ebook_search.search import SearchResult
 
 if TYPE_CHECKING:
@@ -23,6 +24,44 @@ def make_async_client(mocker: MockerFixture, fake_post) -> httpx.AsyncClient:
     return client
 
 
+async def test_model_endpoint_checks_share_http_probe(mocker: MockerFixture) -> None:
+    client = mocker.MagicMock(spec=httpx.AsyncClient)
+    response = mocker.MagicMock(spec=httpx.Response)
+    client.get = mocker.AsyncMock(return_value=response)
+    config = EbookSearchConfig(
+        rerank=RerankConfig(enabled=False),
+        embedding_base_url="https://embedding.example/v1/",
+        vllm_base_url="https://chat.example/v1/",
+        vllm_api_key="secret",
+    )
+
+    assert await check_embedding_endpoint(client, config, timeout_seconds=2.0)
+    assert await check_chat_endpoint(client, config, timeout_seconds=3.0)
+    assert client.get.await_args_list == [
+        mocker.call("https://embedding.example/v1/models", headers={}, timeout=2.0),
+        mocker.call(
+            "https://chat.example/v1/models",
+            headers={"Authorization": "Bearer secret"},
+            timeout=3.0,
+        ),
+    ]
+    assert response.raise_for_status.call_count == 2
+
+
+async def test_model_endpoint_checks_report_http_failures(mocker: MockerFixture) -> None:
+    client = mocker.MagicMock(spec=httpx.AsyncClient)
+    client.get = mocker.AsyncMock(
+        side_effect=[
+            httpx.ConnectError("embedding offline"),
+            httpx.ConnectError("chat offline"),
+        ]
+    )
+    config = EbookSearchConfig(rerank=RerankConfig(enabled=False))
+
+    assert not await check_embedding_endpoint(client, config)
+    assert not await check_chat_endpoint(client, config)
+
+
 async def test_answer_query_uses_httpx_chat_completions(mocker: MockerFixture) -> None:
     captured: dict[str, object] = {}