AI-driven, self-healing web scraping: pluggable render backends + LLM-proposed extraction candidates validated and judged against real page content, never hand-rolled per-site parsers