Fix DAct input ordering of gradient input and activation input

[jberchtold-nvidia]

Description

Fixes an issue where nvte_quantize_dbias_ and cast_fp8_2D where the gradient input and activation inputs were incorrectly swapped in their usages.

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 refactoring

Changes

Please list the changes introduced in this PR:

• Fix internal functions used by nvte_quantize_dbias_ so it can now be used correctly with inputs of gradient input followed by activation input
• Update dact_fn in activation_template.h to undo the swapping of parameters
• Update cast_fp8_2D_kernel to fix usages of input and act_input tensors to correctly usage input as the gradient and act_input as the original activation input from the forward pass
• Add unit tests for DAct where the output is FP8 to test these kernels. Included a test where the input gradient is all zeros to highlight a case that was failing without this change where the output gradient was not all zeroes

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

[jberchtold-nvidia]

/te-ci

[jberchtold-nvidia]

/te-ci

[ptrendx]

Fixed in #1462, which was just merged. Closing this PR.