Enable MXFP8 support in TE JAX integration

[Micky774]

Description

Enables MXFP8 support in the TE JAX integration and significantly modifies tests to account for remaining support gaps

TODO:

• Investigate grouped-GEMM test failures for MXFP8
• Investigate newly exposed failures in test_layernorm_mlp_grad{_shardy} when using hipblaslt GEMM + bias + no scaling + bf16 dtype

Fixes # (issue)

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:

• Adds new test cases to tests/jax/test_custom_call_compute.py
• Corrects comparison bug in test_custom_call_compute::assert_bitwise_scaled_tensors
• Removes dtype-based skip on MXFP8 GEMMs in test_custom_call_compute
• Adds shape-based skips for MXFP8 GEMMs across several files
• Added bias parameterization to test_dense_grad_fp8 to allow testing of MXFP8 (currently does not support bias)
• Updated tests/jax/test_distributed_layernorm_mlp.py with new test shape to allow for MXFP8 usage
• Added xfail to certain configs that fail with the new test case (needs a follow-up investigation)
• Added explicit shape checks to transformer_engine/common/gemm/rocm_gemm.cu
• Removed IS_NORM template parameter from cast_mxfp8_2D_kernel
• Disabled scale_inv swizzling before GEMM on ROCm
• Corrected scale_inv un-padding behavior in NormFwdPrimitive
• Removed redundant un-padding in NormFwdPrimitive
• Removed swizzling in GroupedGemmFFI
• Skipped grouped GEMM MXFP8 tests due to outstanding failures (needs follow-up investigation)
• Corrected bias bug in transformer_engine/jax/layernorm_mlp.py

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

[Micky774]

Note that the CI failure is unrelated.

[wangye805]

There are still many other places needs rocm specific guards

[wangye805]

LGTM