Adding NUMA support for pytorch (#3926)

Summary:

X-link: facebookresearch/FBGEMM#1014

X-link: pytorch/pytorch#150597

Add entry point and environment variable to control NUMA binding / assigment for distributed pytorch runs (training).

Differential Revision: D72321369

Author: efiks

Diff for: fbgemm_gpu/test/sparse/common.py

@@ -16,7 +16,7 @@
 import fbgemm_gpu
 import torch
 from hypothesis import HealthCheck, settings
-from torch._utils_internal import get_file_path_2  # @manual=//caffe2:utils_internal
+from torch._utils_internal import get_file_path_2  # @manual=//caffe2:_utils_internal
 from torch.testing._internal.optests import generate_opcheck_tests
 
 # pyre-fixme[16]: Module `fbgemm_gpu` has no attribute `open_source`.

Diff for: fbgemm_gpu/test/test_utils.py

@@ -141,7 +141,7 @@ def decorator(test_class: unittest.TestCase) -> unittest.TestCase:
             if not has_optests():
                 return test_class
             import torch.testing._internal.optests as optests
-            from torch._utils_internal import (  # @manual=//caffe2:utils_internal
+            from torch._utils_internal import (  # @manual=//caffe2:_utils_internal
                 get_file_path_2,
             )