[TorchArrow][efficiency][2/n] added inference_wrapper_run_flat_out fused version registration call (#81131)
Summary:
Added registration call to make pytorch runtime aware of `inference_wrapper_run_flat_out` functionality. Added fused version of op which will enable out variants when [OptimizeGraph pass](https://fburl.com/code/lsagnmge) is made.
Next: need to add variadic version of fused and unfused op.
Reviewed By: tgalkovskyi
Differential Revision: D37139204
Pull Request resolved: https://github.com/pytorch/pytorch/pull/81131
Approved by: https://github.com/tenpercent, https://github.com/qxy11
