[TE/JAX] Enabling CudaGraph for custom calls with FFI

[phu0ngng]

Description

This PR added traits=1 to the register_custom_call_target(), to enable cudaGraph for the custom calls with FFI.
This was somehow missing in #946.

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 refractor

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

[phu0ngng]

/te-ci jax

[phu0ngng]

It was verified that cudaGraph capture worked for the three custom calls with FFI. Thus the PR is ready.

FYI, one can check if cudaGraph work via nsys, as:

$ XLA_FLAGS=--xla_gpu_graph_min_graph_size=1 nsys profile --cuda-graph-trace=node -o custom-call-graph --force-overwrite=true pytest tests/jax/test_custom_call_compute.py::TestActivationLu::test_activation_lu[activation_type0-shape0]
$ nsys stats -r cuda_kern_exec_trace --force-export=true custom-call-graph.nsys-rep

[nouiz]

LGTM