evaluator_optimizer assumes expert feedback wanted

[bandinopla]

I'm playing with the workflow/evaluator.py and instead of striving for EXCELLENT I'm aiming for POOR.
I've set the generator's instuction to "You are a kid playing with the keyboard"
@fast.evaluator_optimizer also doesn't have the option to strive for mediocrity, it only has min_rating no max rating. But still, when it pass the first generation back to the generator it appends the text You are tasked with improving your previous response based on expert feedback and I never said or wanted an expert. This value is hardcoded :

fast-agent/src/fast_agent/agents/workflow/evaluator_optimizer.py

Line 314 in 4b7e941

You are tasked with improving your previous response based on expert feedback. This is iteration {iteration + 1} of the refinement process.

What if one wants a POORLY redacted message? To simulate a low IQ or dumb person's writing style.

[evalstate]

Would definitely review a PR to allow the evaluator prompt to be customised :)

[bandinopla]

A humble addition: #538