Improve LLM debugging, repair, retry feedback, and CLI output
- Add --debug support to fingerprint_style.py and apply_fingerprint.py for jq-formatted parsed LLM JSON emitted to stderr
- Forward debug and author-name options through shell wrappers, including local db-priv/stylo.sh
- Harden apply_fingerprint JSON parsing with deterministic repair for common malformed quoted strings before falling back to LLM repair
- Strengthen apply prompts to require escaped double quotes in JSON string fields
- Add forced-person pronoun mitigation with deterministic sentence repair and sentence-only LLM repair before full chunk retries
- Add --author-name handling for --3rd-person rewrites so prompts can occasionally use the author name, especially after headings
- Improve style retry feedback with prioritized failed constraints, actual/target measurements, response diagnostics, and preservation instructions
- Add regression coverage for JSON quote repair, author-name third-person prompting, pronoun repair, and actionable style retry feedback
- Redesign CLI output for apply/fingerprint flows with compact colorized box-drawing tables, streaming headers, row separators, wrapped cells, consolidated final result output, and NO_COLOR/CLICOLOR support
- Update README.md and AGENTS.md for new CLI flags, repair/retry behavior, pronoun handling, debug output, and console formatting