[RFE] Establish refinement loop that allows interaction from end user to improve a result

[jwmatthews]

Consider adding the ability for an end user to improve a proposed change via sharing comments on what they want.

Rough idea:

• End user will view the proposed change from the IDE (i.e. accept/reject workflow)
• Assume that the end user wants to see some changes with what was proposed
• User will specify via natural language their desire of what they want to see changed, lets call this a 'change description'
• IDE communicates back to Kai that the proposed change is insufficient, the user wants to see changes as per 'change description'
• Kai iterates with the LLM and improves the proposed change
• Newer version of proposed change is given back to end user
• Potential to loop over this flow and adjust as needed.

[jwmatthews]

@fabianvf please read over and help update if I missed anything from our discussion.

[fabianvf]

Yeah this is what I was picturing. Should be able to do it without much change required Kai side