feat(trtllm): add nsys GPU profiling for TRT-LLM generation backend

[Superjomn]

Wire the TRT-LLM generation backend into NeMo-RL's existing nsys worker-pattern interface (NRL_NSYS_WORKER_PATTERNS / NRL_NSYS_PROFILE_STEP_RANGE).

• The sync/async generation actors resolve nsys options via a TRT-LLM adapter and forward them into TRT-LLM's inner Ray executor through ray_worker_nsight_options (setdefault, so user kwargs win), so the ranks that actually run the engine are profiled.
• The adapter layers TRT-LLM-only options (python-gil,osrt trace, cuda-memory-usage, capture-range-end=repeat-shutdown:1, kill=none, %h in the output name) on top of the shared base config without mutating it, keeping other backends' defaults unchanged.
• The TRT-LLM engine self-manages its capture window, so the trainer's start/stop_gpu_profiling hooks are no-ops for this backend.
• Add focused contract tests and docs.

What does this PR do ?

Add a one line overview of what this PR aims to accomplish.

Issues

List issues that this PR closes (syntax):

Usage

• You can potentially add a usage example below

# Add a code snippet demonstrating how to use this

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you run the unit tests and functional tests locally? Visit our Testing Guide for how to run tests
• Did you add or update any necessary documentation? Visit our Document Development Guide for how to write, build and test the docs.

Additional Information

• ...