Fix: Handle list token in LangChain callback on_llm_new_token to prevent AttributeError

[2429-mori-a]

Background

When using certain models or integrations with LangChain, the on_llm_new_token callback may receive a list of tokens instead of a string.
Chainlit currently expects token to be a string and directly calls string methods such as strip() on it.
If a list is received, this results in an AttributeError ('list' object has no attribute 'strip').

What does this PR do?

• Adds type checking for the token argument in the on_llm_new_token callback.
• If token is a list, its elements are joined into a single string.
• If token is not a string, it is converted to a string.
• This ensures compatibility with all LangChain model integrations and prevents runtime errors when a list is received.

How to test

• Run the existing tests to ensure there are no regressions.
• (Optional) Add a test where a list is passed as the token argument to confirm that no error occurs and the behavior is as expected.

---

Please let me know if any changes or improvements are needed!