add test coverage for loading state dict from cpu for EBC sharding

Summary:
# context
* OSS CI [test is failing](https://github.com/pytorch/torchrec/actions/runs/17565037476/job/49890092618) due to incorrect sharded EBC weights, which further comes from a recent change D80547120, where the eager-mode EBC is on CPU
* the `load_state_dict` is skipped from [day-1](https://www.internalfb.com/diff/D46653763?dst_version_fbid=803682924606242&transaction_fbid=6359829970773670)
* actually this feature was added in T156003280 by D46915880

Differential Revision: D82002643

Author: TroyGarden

torchrec/distributed/embeddingbag.py

@@ -597,10 +597,7 @@ def __init__(
         if env.process_group and dist.get_backend(env.process_group) != "fake":
             self._initialize_torch_state()
 
-        if module.device not in ["meta", "cpu"] and module.device.type not in [
-            "meta",
-            "cpu",
-        ]:
+        if module.device != "meta" and module.device.type != "meta":
             self.load_state_dict(module.state_dict(), strict=False)
 
     @classmethod