PL-135538: Replace real llama-server tests with in-process dummy engine

- Add src/skvaider/dummy_engine/ -- DummyModel with HTTP control API
  (POST /__control/set_response, GET /__control/last_request, POST /__control/reset)
- Add DummyModelConfig to inference config's AnyModelConfig discriminated union
- Wire DummyModel into inference server lifespan (lazy import, isinstance branch)
- Replace OpenAIServerMock with DummyModel-based fixture
- Rename gemma -> gemma_real, add gemma fixture using DummyModel (default)
- Rename embeddinggemma -> embeddinggemma_real for stability test
- Update 6 tests to use dummy engine (test_health, test_metrics, test_proxy,
  test_model_name_case_normalization, test_health_monitoring streaming)
- Add aiohttp dev dependency

Author: elikoga

devenv.nix

@@ -25,6 +25,7 @@
 
   env = {
     PYTHONUNBUFFERED = "1"; # makes output from subprocesses in tests more reliably visible
+    UV_PROJECT_ENVIRONMENT = "${toString ./.devenv/state/venv}";
   };
 
   scripts.run-tests.exec = ''

pyproject.toml

@@ -74,11 +74,11 @@ reportPrivateUsage = "none"
 root = "src/skvaider/tests"
 reportPrivateUsage = "none"
 
-
 [dependency-groups]
 dev = [
     "pytest>=8.4.1",
     "pytest-asyncio>=1.1.0",
     "pytest-cov>=6.2.1",
     "pytest-timeout>=2.4.0",
+    "aiohttp>=3.9.0",
 ]