[Draft][JAX] Add "initialize" XLA stage to remaining TE/JAX primitives

[jberchtold-nvidia]

Description

Follow-up of PR #2219 to apply warmup initialization to the rest of the TE/JAX primitives.

Motivation, identical to previous PR:

If CUDA modules are loaded during any live NCCL kernels, this can cause a hang. When lazy module loading is enabled, the CUDA modules of TE/JAX's custom ops won't be loaded until the XLA "execute" stage, which may have active NCCL kernels.
To prevent this, we register an XLA "initialize" stage for these primitives that does a dry-run of the custom op inside a stream capture that is immediately destroyed to trigger loading of the CUDA modules before the "execute" stage of the XLA program starts.

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

• Add stream capture warmup to the XLA initialize stage of the remaining primitives. This includes quantize, dequantize, softmax, fused attention, gemm, and grouped gemm.

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