[feat]: enabling schedule with requests

[JasonHe-WQ]

1. Issue or feature description

Add requests on GPU resources computation ,to enable more scheduling strategies and more utilization on GPU computation

2. Steps to reproduce the issue

apiVersion: v1
kind: Pod
metadata:
  name: gpu-pod
spec:
  containers:
    - name: ubuntu-container
      image: ubuntu:18.04
      command: ["bash", "-c", "sleep 86400"]
      resources:
        requests:
          nvidia.com/gpucores: 20 # Each vGPU uses 20% of the entire GPU （Optional,Integer)
        limits:
          nvidia.com/gpu: 1 # requesting 1 vGPUs
          nvidia.com/gpumem: 3000 # Each vGPU contains 3000m device memory （Optional,Integer）
          nvidia.com/gpucores: 25 # Each vGPU uses 25% of the entire GPU （Optional,Integer)

Only scheduler codes are needed to be edit, and it will be compatible to the HAMi DRA.

Attention: This feature will leads to QoS level change, original Pods with guaranteed QoS level will not be guaranteed with computation resource once any Pod with requests were scheduled on the same GPU. And if this feature is disabled, nothing will be changed.

3. Information to attach (optional if deemed irrelevant)

[JasonHe-WQ]

Closing due to significant uncertainties in the CUDA driver, such as QoS issues, memory allocation for request-only tasks, and more.

For a closed-source commercial implementation, consider run.ai. Their relevant blog posts are provided below:

Maximize the Potential of Your GPUs: A Guide to Dynamic GPU Fractions & Node-Level Scheduler
Dynamic GPU Memory: Solving the Problem of Inefficient Resource Allocation in Inference Servers