Documentation for DDP-related environment variables

[GuWei007]

📚 The doc issue

I found these environment variables in the PyTorch code. Is there any document that describes the application scenarios?
TORCH_NCCL_BLOCKING_WAIT
TORCH_NCCL_ASYNC_ERROR_HANDLING
TORCH_NCCL_DUMP_ON_TIMEOUT
TORCH_NCCL_DESYNC_DEBUG
TORCH_NCCL_ENABLE_TIMING
TORCH_NCCL_ENABLE_MONITORING
TORCH_NCCL_HEARTBEAT_TIMEOUT_SEC
TORCH_NCCL_TRACE_BUFFER_SIZE
TORCH_NCCL_WAIT_TIMEOUT_DUMP_MILSEC
TORCH_NCCL_COORD_CHECK_MILSEC
TORCH_NCCL_ABORT_IN_DESTROY_PG
TORCH_NCCL_AVOID_RECORD_STREAMS
TORCH_NCCL_USE_TENSOR_REGISTER_ALLOCATOR_HOOK
TORCH_NCCL_DEBUG_INFO_PIPE_FILE
TORCH_NCCL_DEBUG_INFO_TEMP_FILE

Suggest a potential alternative/fix

No response

cc @mrshenli @pritamdamania87 @zhaojuanmao @satgera @gqchen @aazzolini @osalpekar @jiayisuse @H-Huang @kwen2501 @awgu @penguinwu @fegin @XilunWu @wanchaol @fduwjj @wz337 @tianyu-l @wconstab @yf225 @chauhang @d4l3k

[GuWei007]

https://pytorch.org/docs/stable/torch_environment_variables.html
It seems that this document is about environment variables. Can you explain and supplement it in this document? It would be better to give a demo and explain the application scenario.

[awgu]

Closed with #128235

[awgu]

I realized some env vars like TORCH_NCCL_AVOID_RECORD_STREAMS and TORCH_NCCL_USE_TENSOR_REGISTER_ALLOCATOR_HOOK are still not documented (but maybe intentionally?)

[cjxjxjx]

The following environment variables are not documented: TORCH_NCCL_ASYNC_ERROR_HANDLING, TORCH_NCCL_COORD_CHECK_MILSEC, TORCH_NCCL_NONBLOCKING_TIMEOUT, TORCH_NCCL_RETHROW_CUDA_ERRORS, TORCH_NCCL_TRACE_CPP_STACK, TORCH_NCCL_USE_TENSOR_REGISTER_ALLOCATOR_HOOK.

[fduwjj]

Yep, we do need to add more documents around that.