Add GPU support;Use resource limits for resource detection

[chihchun1011]

Summary

This PR makes two changes:

1. Switches resource detection from requests to limits for CPU, memory, and GPU.

2. Adds automatic GPU passthrough. When nvidia.com/gpu is present in limits, the plugin automatically injects the --nv flag into Singularity options to enable NVIDIA runtime support inside the container.

A test manifest (tests/gpu_k8s.yaml) is included to validate GPU end-to-end via nvidia-smi.

Changes

• Plugin: Read CPU, memory, and GPU from resources.limits instead of resources.requests.
• Plugin: Auto-inject --nv into Singularity options when nvidia.com/gpu is detected.
• Test: Add tests/gpu_k8s.yaml — runs nvidia-smi via the Interlink HTCondor virtual node.

Testing

kubectl apply -f tests/gpu_k8s.yaml
# Expected: nvidia-smi lists GPUs, CUDA_VISIBLE_DEVICES=0,1

[dciangot]

Thank you @chihchun1011 , we are almost there with the checks!

Warning: orkspace/tests/gpu_k8s.yaml:1:1: [warning] missing document start "---" (document-start)
Warning: orkspace/tests/gpu_k8s.yaml:8:3: [warning] wrong indentation: expected 4 but found 2 (indentation)
Warning: orkspace/tests/gpu_k8s.yaml:28:3: [warning] wrong indentation: expected 4 but found 2 (indentation)

[dciangot]

LGTM

[dciangot]

I'm approving this; the only left test failure is a flake.