Avoid structural dependency on LLM APIs

[0xba1a]

The current implementation of src/microbots/llm/openai_api.py is heavily depending on the response structure of the LLM. It expects the LLM's response to be always in llm_output_format. This adds an additional overhead for the custom-bots which need to extend the current system-prompt including the specific llm_output_format.

We should converge with OpenAI API's original structure. The LLM function-call should be in dedicated LLM's function-call key instead of forcing the LLM to provide it's NL output in JSON everytime.