fix: improve speaker selection in SelectorGroupChat for weaker models

[ekzhu]

Don't throw an exception when model makes a mistake. Use retries, and if not succeeding after a fixed attempts, fall back to the previous sepaker if available, or the first participant.

Resolves #5453

[afourney]

Looks good. This is a pretty general pattern. We need it in many places like web surfer, orchestrator, file surfer, etc. Really, anywhere where structured or verifiable output is needed. I wonder if there's a more general component or strategy we could provide?

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 78.20%. Comparing base (0b659de) to head (973cf48).
Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #5454      +/-   ##
==========================================
+ Coverage   78.11%   78.20%   +0.08%     
==========================================
  Files         158      158              
  Lines        9602     9626      +24     
==========================================
+ Hits         7501     7528      +27     
+ Misses       2101     2098       -3     

Flag	Coverage Δ
unittests	78.20% <100.00%> (+0.08%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[ekzhu]

Looks good. This is a pretty general pattern. We need it in many places like web surfer, orchestrator, file surfer, etc. Really, anywhere where structured or verifiable output is needed. I wonder if there's a more general component or strategy we could provide?

Yeah, it seems to be a general strategy when it comes to generation-verification. We can consider making this more general, e.g.,

async def RetryOnVerification(model_client, initial_prompt, verification_func, default_output):
    ...