feat: warn on import for unsupported torchao/torch pairs

Read torchao's version via importlib.metadata rather than
torchao.__version__: on torch < 2.10, importing torchao >= 0.18.0
raises ImportError, so the warning must be emitted before the
imports below reach torchao.

Author: guru-desh

src/coreai_opt/__init__.py

@@ -8,9 +8,24 @@
 For deployment via Core AI on Apple Silicon.
 """
 
-from . import palettization, pruning, quantization
-from ._about import __version__
-from .common import CoreMLExportError, ExportBackend
+import importlib.metadata
+import warnings
+
+import torch
+
+from coreai_opt._utils.version_utils import (
+    torchao_torch_incompatibility as _torchao_torch_incompatibility,
+)
+
+_incompatibility = _torchao_torch_incompatibility(
+    importlib.metadata.version("torchao"), torch.__version__
+)
+if _incompatibility:
+    warnings.warn(_incompatibility, UserWarning, stacklevel=2)
+
+from . import palettization, pruning, quantization  # noqa: E402
+from ._about import __version__  # noqa: E402
+from .common import CoreMLExportError, ExportBackend  # noqa: E402
 
 __all__ = [
     "CoreMLExportError",