Add sharding optimization feature rollout (pytorch#2680)

Summary:
Pull Request resolved: pytorch#2680

As per title, adding JK to enable gradual rollout of sharding optmization in torchrec per entitlement
https://www.internalfb.com/intern/justknobs/?name=ai_infra%2Ftraining_components#torchrec_sharding_optimization_enable

Reviewed By: lijia19, iamzainhuda

Differential Revision: D68038832

fbshipit-source-id: 29fa81cedea8d87044b5fc35238627b6cd327e0c

torchrec/distributed/global_settings.py

@@ -30,3 +30,7 @@ def construct_sharded_tensor_from_metadata_enabled() -> bool:
     return (
         os.environ.get(TORCHREC_CONSTRUCT_SHARDED_TENSOR_FROM_METADATA_ENV, "0") == "1"
     )
+
+
+def enable_construct_sharded_tensor_from_metadata() -> None:
+    os.environ[TORCHREC_CONSTRUCT_SHARDED_TENSOR_FROM_METADATA_ENV] = "1"