Fix issue #385: Decouple Quark quantization via Docker and resolve flexml dependency conflicts

[zeenat28-ui]

Summary of Changes

This PR resolves Issue #385 where running pip install -r requirements.txt inside the Ryzen AI 1.8 virtual environment fails due to dependency conflicts between AMD Quark/torchvision and flexml 1.8.0 (torch==2.6.0+cpu, torchvision==0.21.0+cpu, onnx==1.21.0).

Key Fixes:

1. Decoupled Build/Quantization vs. Runtime Environments:
   • Created Dockerfile and docker_run.sh / docker_run.bat to run Quark INT8 quantization in an isolated container without polluting the host environment.
   • Separated requirements into requirements_quantize.txt (for Quark quantization) and requirements_infer.txt (minimal packages for NPU runtime without conflicting torch/torchvision pins).
2. Git-LFS Checkpoint Handling in prepare_model_data.py:
   • Added automatic download of the full 90.5 MB pre-trained ResNet-50 checkpoint from Git-LFS if only the 133-byte pointer is present.
3. Comprehensive Documentation:
   • Updated Readme.md with step-by-step instructions for both Docker and local dual-conda workflows.