[Core][MM] Optimize encoder cache manager by operating with embeddings only (#30475)

Signed-off-by: Roger Wang <hey@rogerw.io>
Co-authored-by: Sun Kim <sunytokki@gmail.com>
(cherry picked from commit f5f51e5931)

tests/v1/ec_connector/unit/test_ec_example_connector.py

@@ -38,7 +38,7 @@ class MockRequest:
             )
             self.mm_features.append(feature)
 
-    def get_num_encoder_tokens(self, input_id: int) -> int:
+    def get_num_encoder_embeds(self, input_id: int) -> int:
         assert input_id < len(self._token_counts)
         return self._token_counts[input_id]