Improve experience for poor generateRebase responses

[sergeibbb]

Description

Poor responses

We should have an evaluation solution to help determine poor responses before the user does and use another turn to retry and improve the quality of the response.

[sergeibbb]

@d13

How do you see the approach?

1. Launch independent AI conversation to evaluate the solution?
2. Assuming that every commit needs to be compilable, lintable, to have unit-test passing. Therefore do we want to checkout to every generated commit and launch the compilation and unit test checking?
3. Did you mean anything else?

Is there an example of a poor response IRL that we want our tool to be able to improve?