fix: correct role assignment in chat_history and typo in get_embedding

[dajiaohuang]

What

Two fixes:

1. chat_history.py: In to_messages(), messages from other agents (i.e., not my_name and not "function") were incorrectly assigned "assistant" role. These should be "user" role to maintain proper conversation alternation required by models like LLAMA that enforce user/assistant/user alternation.

2. openai.py: Fixed NameError in get_embedding() exception handler where attempt was used instead of attempts (the function parameter).

Why

1. Issue why the to_message function in chat_history.py is trying to add an assistent message? #133: When using LLAMA models via vLLM, the API returns BadRequestError: Conversation roles must alternate user/assistant/user/assistant/... because messages from other agents in multi-agent conversations were being labeled as "assistant" role.

2. The get_embedding function has a bug where if an exception occurs, it tries to increment an undefined variable attempt instead of the parameter attempts.

Testing

• The chat_history fix follows the same pattern already used in the codebase where my_name messages are labeled "assistant" and "function" messages are labeled "function" - messages from other agents should logically be "user".
• The get_embedding fix is a simple typo correction.