CI: frequency of hitting timeout/network errors has significantly increased recently

[leofang]

This can happen during

• pip install
  • Ex: CI: Windows GPU runners do not stop on error #483
• fetching artifacts from GitHub
  • Ex: https://github.com/NVIDIA/cuda-python/actions/runs/13623473149/job/38077154585#step:10:219

[leofang]

xref: https://github.com/NVIDIA/cuda-python/actions/runs/14048031704?pr=517

[rwgk]

xref: https://github.com/NVIDIA/cuda-python/actions/runs/14087083558/job/39461464660?pr=503

It took 4 reruns until all tests passed.

The current situation is quite disruptive, especially if I need to weed out real failures. The general issues are akin to decoys.

[leofang]

We've observed no more network issues lately! According to @ajschmidt8:

Most likely moving the V100s from RDS Lab to NVKS resolved the network issues.
The NVKS cluster is in a different networking environment that seems much more stable than RDS Lab. Hopefully it stays that way!