[PyTorch|common] Optimize unpadding kernel for FP8

[xiaoxi-wangfj]

1. Add multi-tensor unpadding kernel
2. Replace split+cat with unpadding kernel in Fp8Padding and Fp8Unpadding
3. Add unpadding with padding unit tests

Description

This PR introduces a high-performance CUDA kernel implementation for tensor unpadding, replacing the previous inefficient torch.split + torch.cat approach. Key improvements include:

• ​​2x faster unpadding kernel performance​​ (measured with microbenchmarks).
  ​​- 0.3% MFU uplift​​ in my end-to-end 1000B parameter model training.
• Added comprehensive unit tests for edge cases (e.g., partial padding units).

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)
• [x ] 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 refactoring

Changes

• Replace the existing torch.split + torch.cat unpadding implementation in Fp8Padding backward pass and Fp8Unpadding forward pass
• Add unpadding inplement in common
• Add unpadding unit test in tests
  ​​

Checklist:

• [ x] I have read and followed the contributing guidelines
• [ x] 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
• [x ] 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

[yaox12]

/te-ci

[yaox12]

/te-ci