enable feature score auto collection in EBC (#5031)

Summary:

X-link: meta-pytorch/torchrec#3475

X-link: facebookresearch/FBGEMM#2044

Enable feature score auto collection for EBC in the similar way of EC. The configuration has no difference in embedding table config:

              virtual_table_eviction_policy=FeatureScoreBasedEvictionPolicy(
                  training_id_eviction_trigger_count=260_000_000,  # 260M
                  training_id_keep_count=160_000_000,  # 160M
                  enable_auto_feature_score_collection=True,
                  feature_score_mapping={
                      "sparse_public_original_content_creator": 1.0,
                  },
                  feature_score_default_value=0.5,
              ),

Reviewed By: EddyLXJ

Differential Revision: D85017179

Author: emlin

fbgemm_gpu/fbgemm_gpu/split_table_batched_embeddings_ops_common.py

@@ -241,6 +241,7 @@ class KVZCHParams(NamedTuple):
     backend_return_whole_row: bool = False
     eviction_policy: EvictionPolicy = EvictionPolicy()
     embedding_cache_mode: bool = False
+    feature_score_collection_enabled: bool = False
 
     def validate(self) -> None:
         assert len(self.bucket_offsets) == len(self.bucket_sizes), (