Explore LLM-as-a-Meta-Judge self improving Judge loop

[LeoVS09]

Meta introduced new approach of unsupervised Judge agent self-improvement. Probably it can be useful to increase quality of SADD and SDD judges, but may require continues learning strategy implementation.