[PyTorch] Remove some direct calls to PyTorch extensions in Float8Tensor

[timmoon10]

Description

When Float8Tensor launches TE kernels, it directly calls the PyTorch extensions rather than using the Python wrapper functions in te.pytorch.cpp_extensions. This is because the Python wrapper functions required passing in FP8TensorMeta objects. #1083 changed the APIs so the FP8TensorMetas are optional. This PR takes advantage of these new APIs, which also has the effect of registering FP8 casts with TorchScript.

Type of change

• Documentation change (change only to the documentation, either a fix or a new content)
• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Infra/Build change
• Code refractor

Changes

• Replace tex.cast_to_fp8 and tex.cast_from_fp8 calls in Float8Tensor with cpp_extensions wrappers

Checklist:

• I have read and followed the contributing guidelines
• The functionality is complete
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes

[timmoon10]

/te-ci pytorch

[timmoon10]

These changes have been incorporated into #1127.