fix: pynvml shutdown error

[Nexisato]

Description

• fix: pynvml shutdown exception

Demo log exception

• URL: https://dev.powerducker.com/@nexisato/4-gpu-nccl-test/runs/0912e08a-450a-4c9b-bdae-e4845ba856fc/log
• Logs:

 Exception ignored in: <function GpuCollector.__del__ at 0x7f34a2572d40>
Traceback (most recent call last):
  File "/usr/local/miniconda3/envs/py312/lib/python3.12/site-packages/swanlab/data/run/metadata/hardware/gpu/nvidia.py", line 267, in __del__
    pynvml.nvmlShutdown()
  File "/usr/local/miniconda3/envs/py312/lib/python3.12/site-packages/pynvml.py", line 2970, in nvmlShutdown
    _nvmlCheckReturn(ret)
  File "/usr/local/miniconda3/envs/py312/lib/python3.12/site-packages/pynvml.py", line 1083, in _nvmlCheckReturn
    raise NVMLError(ret) 

Issues: #1532

[gemini-code-assist]

Code Review

This pull request adds error handling to NVML shutdown calls in the GPU collector to prevent crashes. The review suggests changing the warning log in get_nvidia_gpu_info to a debug log to reduce noise and recommends silencing exceptions in the del method to avoid potential AttributeError issues during interpreter shutdown.

[Nexisato]

/gemini review

[gemini-code-assist]

Code Review

This pull request implements a cross-platform file-locking mechanism to prevent redundant hardware monitoring in multi-process environments, specifically supporting a new parallel="shared" mode. Changes include the addition of msvcrt and fcntl logic for lock management, updates to the SwanLabRun lifecycle for lock acquisition and release, and improved error handling during GPU library shutdown. Review feedback suggests using more idiomatic if statements instead of logical operators for side effects, avoiding the deletion of advisory lock files to prevent potential race conditions, and cleaning up unused exception variables.

[Nexisato]

有多物理节点问题，该方案暂时不适用