refactor: unify CUDA skip guards in tests

- add skip_no_fp8 (CUDA + fp8_mm kernel + cc 8.9+) to tests/conftest.py
- use skip_no_cuda / skip_no_kernel / skip_no_fp8 directly in test modules
- drop _GPU alias and tests.extension.conftest re-exports
- remove unused imports (Union in hf_adapter, make_grpo_config in data conftest)

astrai/serialization/hf_adapter.py

@@ -21,7 +21,7 @@ Not supported:
 
 import logging
 import re
-from typing import Any, Dict, Mapping, Union
+from typing import Any, Dict, Mapping
 
 import torch