Add github-base-research experimental skill

[adamsardo]

Summary

• add a new experimental skill: github-base-research
• provide a parallel GitHub search workflow for identifying reusable base repositories
• include scoring/ranking that recommends either a single base repo or a composable multi-repo foundation
• add supporting references for query strategy, evaluation rubric, and report format
• incorporate field feedback to improve reliability on macOS and under GitHub API burst limits

Improvements from real usage feedback

• remove Bash-4-only readarray usage so collector works on macOS default Bash 3.2
• add API retry/backoff for rate-limit and secondary-limit responses
• add serial fallback when parallel collection/hydration hits burst limits
• add collector filters for signal quality (--min-stars, archived/fork filtering, trimmed metadata output)
• add ranking hard-gate filtering (--min-stars, --max-inactive-days, optional --require-license, --no-hard-gate)
• update playbook/docs with query qualifiers and safe-mode collection guidance

Files

• skills/.experimental/github-base-research/SKILL.md
• skills/.experimental/github-base-research/agents/openai.yaml
• skills/.experimental/github-base-research/scripts/collect_github_repos.sh
• skills/.experimental/github-base-research/scripts/rank_github_bases.py
• skills/.experimental/github-base-research/references/query-playbook.md
• skills/.experimental/github-base-research/references/evaluation-rubric.md
• skills/.experimental/github-base-research/references/report-template.md
• skills/.experimental/github-base-research/LICENSE.txt

Validation

• python3 skills/.system/skill-creator/scripts/quick_validate.py skills/.experimental/github-base-research
• bash -n skills/.experimental/github-base-research/scripts/collect_github_repos.sh
• python3 -m py_compile skills/.experimental/github-base-research/scripts/rank_github_bases.py
• run collect_github_repos.sh with a focused query and --no-hydrate to verify end-to-end collection
• run rank_github_bases.py on fixture input to verify ranking/report output