feature request: Provide better callback variables.

[tusharhero]

The current callback variable ellama-chat-done-callback, as the name
implies, only calls the function after it has finished processing the
entire response. This means there is no support for streaming
functionality. I have a text-to-speech function attached, but I need
to wait a long time before I can listen to the response due to this.

Additionally, in reasoning models, there is a lot of content within
the think block (which appears as a quote block in Org mode). It
must be determined whether we should treat this differently or add an
identifier to indicate that it is a "think" block. Having the option
to completely ignore such blocks when calling the callback function
would be ideal.