Update the langchain/langgraph `on_llm_start` hook to use a different `op`

At the moment, the `op`is set to `gen_ai.pipeline` which is a legacy value that we have deprecated (and only appears in the Python SDK).

We need to update this to something more appropriate, either a text completion or chat completion. Which thing we update to will be determined by the outcome of [this RFC](<https://github.com/getsentry/rfcs/pull/153>). 