To resolve test issues and deploy issues
In the future, we can install pillow by default, and then move back
LMM agents into agentchat

To resolve dependencies issue for build

So, it can still handle human_input_mode
and max_consecutive_reply

Fixing: typos, blogs, yml, and add OpenAIWrapper

Reference  https://platform.openai.com/docs/guides/vision

1. Add a notebook for GPT-4V. LLava notebook also updated.
2. img_utils updated
3. GPT-4V formatter now return base64 image with mime type
4. Infer mime type directly from b64 image content (while loading
   without suffix)
5. Test cases modified according to all the related changes.