[common] Remove kvpacked and qkvpacked attention functions for every kernel type.

[pggPL]

Description

There are 3 variants of fused_attention functions: for separate QKV, KV packed and QKV packed, which differ only by pointers to qkv. This results in code duplication for each type of the fused attention kernel: arbitrary seqlen, max 512 and fp8. This PR deduplicates the code and moves pointer computation one abstraction layer - from the functions like fused_attn_max_512_fwd_qkvpacked into the functions like nvte_fused_attn_fwd_qkvpacked in common c++ api.

These packed versions of common attention api functions are used by JAX, so I think running JAX CI is good test of that changes. PyTorch uses only non-packed function.

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

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

[pggPL]

/te-ci jax

[greptile-apps]

_{7 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[cyanguwa]

I think this is similar to #2272 :) Yes, Jax needs a bit of fixing in order to get its attention working.