LangChain completion wrappers drop runtime kwargs

Describe the bug

Both LangChainLLMCompletionFn.__call__() and LangChainChatModelCompletionFn.__call__() accept runtime **kwargs, but discard them when invoking the wrapped LangChain model:

response = self.llm(prompt)
...
response = self.llm(messages).content

The shared CompletionFn contract documents these kwargs as arguments passed through to the underlying API. Options such as stop, callbacks, or model-supported invocation controls therefore silently disappear when a LangChain wrapper is used.

Expected behavior

Invocation-time kwargs accepted by the completion wrapper should be forwarded unchanged to the wrapped LangChain LLM/chat model. Calls with no kwargs should behave exactly as before.

Proposed fix

Pass **kwargs to both wrapped model calls and add local recording-model tests for the text and chat wrappers.