fix local pytorch build

[dolpm]

Stack from ghstack (oldest at bottom):

• [torchcomms] support regional inductor for collectives #120
• [bc] use coalesce api with backendwrapper #118
• use coalesce api in device mesh functional backend #117
• fullgraph compiler integration -- coalesced collectives #116
• coalesced collective support #115
• dtensor integration testing #114
• device mesh / funcol integration #113
• [torchcomms] add autograd #112
• [torchcomms] torch.compile post-split #111
• add torch compile debug to gitignore #110
• [torchcomms] param parsing utils #109
• [torchcomms] object pickling #108
• -> fix local pytorch build #107
• build_ncclx b200 #119