Remove torch._running_with_deploy() from fbcode and add RuntimeError handling for load_library (#3207)

Summary:
Pull Request resolved: #3207

As per https://fb.workplace.com/groups/pytorch.dev/permalink/1828123831099422 we can now safely remove “torch.is_deploy_running”. This commit does this!

Additionally we are handling RuntimeErrors while using load_library in torchrec

Reviewed By: nipung90

Differential Revision: D78525065

fbshipit-source-id: 85e76dc6ced60c5b286153c2156def2c21f7d302

Author: PaliC

torchrec/distributed/train_pipeline/tracing.py

@@ -12,7 +12,6 @@
 from typing import Any, Dict, List, Optional, Set, Tuple, Union
 
 import torch
-
 from torch.distributed._composable.fsdp.fully_shard import FSDPModule as FSDP2
 
 from torch.distributed.fsdp import FullyShardedDataParallel as FSDP