Add support to run distributed training tests in CI

[Nic-Ma]

Is your feature request related to a problem? Please describe.
Now we already have some distributed training test cases, like: https://github.com/Project-MONAI/MONAI/blob/master/tests/test_handler_rocauc_dist.py
Need to be executed in our CI system.
We can run with 2 GPUs in 1 node first.

[IsaacYangSLA]

This seems straightforward. The CI job running inside the container can see all gpus. So we can either add one item in the bash script, or run it as a separate step in CI. I will try to include it.

[wyli]

just a follow-up on this, the CI environment already provides 2 GPUs per job, the remaining tasks are to

• update runtests.sh to run commands such as

  MONAI/tests/test_handler_rocauc_dist.py

  Lines 41 to 44 in 572c3f7

  	# python -m torch.distributed.launch --nproc_per_node=NUM_GPUS_PER_NODE
  	# --nnodes=NUM_NODES --node_rank=INDEX_CURRENT_NODE
  	# --master_addr="192.168.1.1" --master_port=1234
  	# test_handler_rocauc_dist.py

  
  if multiple gpus are detected locally
• double-check that the CI pipeline runs the updated runtests.sh properly