test: isolate chat mocks and cancellation timing

tests/fixtures/mock_injector.py

@@ -3,7 +3,7 @@ from typing import Any
 from unittest.mock import MagicMock
 
 import pytest
-from injector import Provider, ScopeDecorator, singleton
+from injector import Provider, ScopeDecorator, SingletonScope, singleton
 
 from private_gpt.di import create_application_injector
 from private_gpt.settings.settings import Settings, unsafe_settings
@@ -25,6 +25,7 @@ class MockInjector:
         if mock is None:
             mock = MagicMock()
         self.test_injector.binder.bind(interface, to=mock, scope=scope)
+        self.test_injector.get(SingletonScope)._context.pop(interface, None)
         return mock  # type: ignore
 
     def bind_settings(self, settings: dict[str, Any]) -> Settings:

tests/server/chat/anthropic/test_anthropic_client.py

@@ -692,6 +692,7 @@ def test_messages_count_tokens_endpoint_forwards_message_input_and_options(
     test_client: TestClient,
     httpx_mock: HTTPXMock,
 ) -> None:
+    injector.get(ChatService)
     chat_service = Mock()
     chat_service.count_tokens = AsyncMock(
         return_value=CountTokensOutput(input_tokens=11)

tests/server/chat/test_chat_routes.py

@@ -2233,7 +2233,7 @@ async def test_chat_cancels_llm_astream_on_client_disconnection(
         await request_task
 
     try:
-        await asyncio.wait_for(llm_generator_closed.wait(), timeout=2.0)
+        await asyncio.wait_for(llm_generator_closed.wait(), timeout=5.0)
     except TimeoutError:
         pytest.fail(
             "LLM astream_chat_with_tools generator was not closed after client disconnection"