Supporting MultiModal Models: an example with LLaVA Notebook (#286)

* LMM notebook

* Use "register_reply" instead.

* Loop check LLaVA non-empty response

* Run notebook

* Make the llava_call function more flexible

* Include API for LLaVA from Replicate

* LLaVA data format update x2
1. prompt formater function
2. conversation format with SEP

* Coding example added

* Rename "ImageAgent" -> "LLaVAAgent"

* Docstring and comments updates

* Debug notebook: Remote LLaVA tested

* Example 1: remove system message

* MultimodalConversableAgent added

* Add gpt4v_formatter

* LLaVA: update example 1

* LLaVA: Add link to "Table of Content"