Prepare a pytorch docker used for building xformers on ROCm (AMD MI2xx/MI3xx GPU environment)

[qianfengz]

1. Using latest rocm pytorch nightly docker

   #> docker pull rocm/pytorch-nightly:latest                                                        

2. Create a docker according to your intended torch/compiler/python versions

   1. Get a base ubuntu 20.04 docker 
   2. Install python3.9  
     #> pip install python3.9 python3.9-venv python3.9-dev
   3. Enable python3.9
      #> update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.9
      #> update-alternatives --install /usr/bin/python python /usr/bin/python3
   4. Create venv directory
      #> python -m venv /opt/venv/
   5. Download and install wheels
       * From http://download.pytorch.org/whl/nightly/pytorch-triton-rocm and  http://download.pytorch.org/whl/nightly/torch/
       * pytorch-triton-rocm               ;  wheel which required by torch
       * torch                                       ;  wheel which support rocm6.2+pytorch2.5+py3.9
   6. Save/tag and push the docker for distribution 
   7. When using, enter the venv by  
      #> source /opt/venv/bin/activate