[Feature]support auto nsys capture skill

[rain7996]

Summary
Support an auto nsys capture skill. Include:

• Add a new Claude Code skill (ernie5-nsys-capture) that automates Nsight Systems (nsys) GPU profiling for FastDeploy inference
  services
• The skill provides a 5-step workflow: parameter collection from user's launch script, automatic nvprof_start/nvprof_stop
  injection into gpu_model_runner.py, nsys-wrapped launch script generation, user confirmation, and automated capture execution
• Includes helper shell scripts (nsys_capture.sh for service health check & file management, ernie5_nsys_capture.sh for request
  timing), a default OpenAI-compatible streaming test client, and supports both coarse/detailed nsys profiling modes
• Add a lightweight NVTX range utility (fastdeploy/usage/nvtx.py) that gracefully degrades to a no-op when nvtx is not installed

[paddle-bot]

Thanks for your contribution!

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

songyuxing seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}