[Transform] [Utils] Support precision, add torch dtype validation

[kylesayrs]

Purpose

• Support configuring the precision at which transforms are applied, which seems to have some minor effects on results
• Add TorchDtype type annotation for adding torch dtypes to model definitions

Changes

• Added precision argument to TransformSchemes
  • Transform weights are constructed using this precision
  • Transform weights are applied using this precision
    • This precision is used for both fusing operations and online transforms
• Added TorchDtype type annotation in src/utils/type.py
  • Supports loading from torch.xxx or xxx strings and torch.dtypes

Testing

• Added tests for TorchDtype type annotation
• Tested with different precisions and found torch.float32 to be acceptable

[dsikka]

just a few questions - lgtm otherwise

[brian-dellabetta]

Thanks for the updates to merged precision!