Centralize repository standards in devtools

[artiphishle]

Summary

Implement #24 by making @ankhorage/devtools the canonical owner of shared repository development standards.

What changed

• reorganize implementation around equal concerns under src/tools:
  • eslint
  • prettier
  • knip
  • workflows
  • vscode
• keep all command routing and executable entrypoints under src/cli
• add the canonical command surface:
  • ankh devtools sync [path]
  • ankh devtools status [path]
  • ankh devtools workflows sync|status [path]
  • ankh devtools vscode sync|status [path]
  • --dry-run for all sync commands
• add deterministic, idempotent managed-file synchronization
• preserve unrelated and unknown repository files
• package canonical workflow and VS Code assets with the npm package
• add and dogfood canonical .vscode/settings.json and .vscode/extensions.json
• intentionally leave launch.json unmanaged
• add a minor changeset

Canonical workflows

The managed CI workflow now:

• installs with bun install --frozen-lockfile
• runs bunx @ankhorage/ankh doctor validate . immediately after installation
• conditionally runs build, lint, format check, Knip, tests, typecheck, and pull-request changeset status

The managed release workflow keeps the preferred Bun/Node/Changesets process, concurrency protection, optional build, and graceful skip without a Changesets config.

Architecture and compatibility

• no generic templates directory
• no ankh dev ... alias
• no legacy exports
• no compatibility wrappers
• no migration shims
• no deprecation warnings
• no compatibility with the separate @ankhorage/dev provider

The only canonical provider prefix is ankh devtools ....

Validation

GitHub CI passes completely:

• Ankhorage Doctor validation
• build
• ESLint
• Prettier format check
• Knip
• Bun tests
• TypeScript typecheck
• Changesets status

The tests cover missing/outdated/current files, dry-run behavior, scoped synchronization, preservation of unrelated files, invalid targets, provider routing, packaged asset contracts, and repeated-sync idempotence.

Closes #24