ProcessGroupBaby: support streams in futures

[d4l3k]

There's an issue with the recent future event work that sometimes causes cudart to crash.

This tracks the stream so the future completes in the same stream that it was launched in which seems to avoid the cudart crash.

Stack trace:

Fatal Python error: Segmentation fault

Current thread 0x00007f3752b5c640 (most recent call first):
  File "/home/tristanr/.conda/envs/torchft-3.10/lib/python3.10/site-packages/torch/cuda/streams.py", line 196 in wait
  File "/home/tristanr/torchft/torchft/process_group.py", line 1180 in _future_handler
  File "/home/tristanr/.conda/envs/torchft-3.10/lib/python3.10/threading.py", line 953 in run
  File "/home/tristanr/.conda/envs/torchft-3.10/lib/python3.10/threading.py", line 1016 in _bootstrap_inner
  File "/home/tristanr/.conda/envs/torchft-3.10/lib/python3.10/threading.py", line 973 in _bootstrap

Test plan:

pytest torchft/process_group_test.py -s -v -x

[fegin]

LGTM