Add `skillfile doctor` to diagnose common setup problems

[eljulians]

Every good CLI has a doctor command (brew doctor, flutter doctor). Right now if something's off with your setup (missing token, stale cache, git not on PATH) you only find out when a command fails with a cryptic error. A doctor command that checks everything upfront would save a lot of headaches.

$ skillfile doctor
  ✓ Skillfile found
  ✓ Skillfile.lock found (8 entries)
  ✗ GitHub token not configured (60 req/hr limit)
    → Run: skillfile init (or set GITHUB_TOKEN)
  ✓ git found on PATH (needed for resolve)
  ✓ Install targets: claude-code (global), cursor (local)
  ✗ cursor not detected on system
    → Cursor skills dir ~/.cursor/skills/ doesn't exist
  ✓ .skillfile/cache/ exists (8 entries cached)
  ✗ 2 entries missing from cache
    → Run: skillfile sync
  ✓ No active conflicts

Things it should check:

• Skillfile + lock file exist
• GitHub token configured (env var, gh CLI, or config file)
• git on PATH (needed for resolve)
• Install targets point to valid platforms
• Target directories actually exist on disk
• Cache state (entries cached, any missing)
• No stale conflict state (.skillfile/conflict)
• Patch files are well-formed

Pure read-only, no network calls, no side effects. Just diagnostics + actionable suggestions when something's wrong.

Implementation lives in cli/src/commands/doctor.rs. Most of the building blocks already exist: token detection (sources/src/http.rs), adapter lookup (deploy/src/adapter.rs), lock/cache reading (core/src/lock.rs). It's mostly wiring them together and formatting the output