Refactor `copilot-instructions.md` to reuse repo agent instructions and skills

[ptr727]

Copilot now has support for repo agents and skills, see:

• [https://docs.github.com/en/copilot/concepts/agents/code-review]
• [https://docs.github.com/en/copilot/how-tos/copilot-on-github/use-copilot-agents/copilot-code-review]
• [https://docs.github.com/en/copilot/how-tos/copilot-on-github/customize-copilot/customize-cloud-agent/add-skills]

Refer to for background:

• [https://github.com/ptr727/ProjectTemplate/blob/main/docs/token-cost.md]
• [https://github.com/ptr727/ProjectTemplate/blob/main/docs/token-efficiency-plan.md]
• [https://github.com/Expand references/ in large skill files for progressive disclosure (#766 Phase 2) #773]
• [https://github.com/Optimize Agent Instructions and Skills for Token Efficiency #766]

Implement:

• Create refine / improve code review specific skills, with general and per-language specifics.
• Break up copilot-instructions.md to remove redundancy and duplication and defer to the repo or hub instructions.
• copilot-instructions.md for code review should bootstrap the standard set of code review agents and skills.
• Refer to the several open issues tagged with skills or agent labels that are related to agents ignoring or not following skills and instructions (opencode and some specific models are particularly troublesome), address any obvious issues as they are discovered.
• Explore and evaluate a more programmatic way to interact with copilot reviews:
  • Currently interaction is gated through [https://github.com/ptr727/ProjectTemplate/blob/main/scripts/pr_review.py] to attempt to isolate agents from the seemingly ever-changing copilot review behavior.
  • Improve the script coverage and operations if any new / alternate computer parse-able interaction model is available.
  • Alternatively, or in conjunction with, give copilot instructions to present findings in a machine parse-able format.
  • Instruct copilot to never suppress any findings (suppressed findings do not currently support threaded replies and require free form quoted replies) as history shows 90+% of all suppressed findings are valid and actionable.

Notes:

• The skills and instructions need to be discoverable and supportive of copilot, codex, claude, and opencode.
• There is a skill deployment script that copies the skills for claude discovery as claude does not (to my current knowledge) support general .agents/skills discovery, and creates the skill marketplace json.
• Skills are installed system wide using the host-setup install-skills script.