fix: llm caching for replicate (#6396)

Caching wasn't accounting for which model was used so a result for the
first executed model would return for the same prompt on a different
model.

This was because `Replicate._identifying_params` did not include the
`model` parameter.

FYI
- @cbh123
- @hwchase17
- @agola11

langchain/llms/replicate.py

@@ -72,6 +72,7 @@ class Replicate(LLM):
     def _identifying_params(self) -> Mapping[str, Any]:
         """Get the identifying parameters."""
         return {
+            "model": self.model,
             **{"model_kwargs": self.model_kwargs},
         }