Add Copilot agent skills and expand contributor instructions

[hanniavalera]

Summary

This PR improves the Copilot agent experience for contributors to CMake Tools by:

1. Adding auto-discoverable skills — Step-by-step recipes the agent automatically loads based on task context
2. Expanding contributor instructions — More conventions, patterns, and checklists in .github/copilot-instructions.md
3. Enabling Copilot CI — A setup workflow that bypasses the private Azure Artifacts registry so the agent can build and test

New Skills

Skill	Triggers on	Purpose
add-command	"add command", "register command", extension.ts changes	Adding a new cmake.* command
add-setting	"add setting", "new configuration", config.ts changes	Adding a new cmake.* setting
add-diagnostic-parser	"add parser", "diagnostic parser", src/diagnostics/ changes	Adding a compiler output parser
presets-vs-kits	"presets vs kits", "useCMakePresets", dual-mode code	Writing code that handles both operating modes
backend-tests	"backend test", "unit test", test/unit-tests/backend/ changes	Writing fast Mocha tests without VS Code

Skills use YAML frontmatter with name and description so the agent can auto-load them when relevant.

Expanded Contributor Instructions

New sections in .github/copilot-instructions.md:

• Reactive configuration subscriptions — Use ConfigurationReader.onChange(), not vscode.workspace.onDidChangeConfiguration
• SpecialKits sentinel values — Always check for __unspec__, __scanforkits__ before treating a kit as a real compiler
• Kit trust model — Don't bypass isTrusted filtering
• NLS key naming — Conventions for package.nls.json keys
• Setting scope — When to use "resource" vs "window"
• Adding diagnostic parsers — Quick checklist (detailed recipe in skill)
• Copilot CI environment — Warning about .npmrc.bak and patched yarn.lock
• Backend tests — Import strategy, mirror pattern for vscode-dependent modules

Copilot agent CI environment (agent-only)

This section applies only to the Copilot coding agent, not human contributors.

The Copilot agent runs in a sandbox without Azure Artifacts credentials. The copilot-setup-steps.yml workflow patches the registry configuration to allow yarn install to succeed.

If you are the Copilot agent and git status shows .npmrc.bak or modified yarn.lock:

• These are environment artifacts — do not commit them
• Run git checkout -- .npmrc yarn.lock before committing

If you are a human contributor: Ignore this section. Your local environment should use the standard .npmrc and yarn.lock as-is. If you have Azure Artifacts access issues, see the repo's onboarding docs or contact the team.

Why This Matters

• Faster onboarding — New contributors (human or AI) get actionable guidance without reading the entire codebase
• Fewer review cycles — Skills encode common patterns and checklists, reducing "you forgot X" comments
• Auto-discovery — The agent loads relevant skills based on task context, not just when explicitly invoked
• CI validation — The agent can now run backend tests to catch regressions before opening a PR

Testing

• Skills render correctly in GitHub UI
• yarn compile passes
• yarn backendTests passes
• Copilot agent can load skills based on description triggers (manual testing)

---

No user-facing changes — this is contributor/agent tooling only.