feat(registry): unified CheckRegistry and granular rule execution (tff#146)

[tjirab]

🚀 What changed?

• Introduced centralized declarative CheckRegistry and CheckDefinition in packages/tff-core/src/tff/core/registry.py defining typed metadata (id, label, category, scope, default_severity, aliases, finding_check_id, lazy dynamic loader for rule classes / collector functions, and is_enabled_fn).
• Replaced duplicated hardcoded mapping dictionaries across health.py (CATEGORIES, is_check_enabled, PROJECT_LEVEL_CHECKS) and report.py (CHECK_LABELS, CONNASCENCE_CATEGORIES, ARCHITECTURAL_CHECKS) with single-source-of-truth registry queries.
• Refactored run_all_checks across dbt (tff.dbt.runner), Dataform (tff.dataform.runner), and SQLMesh (tff.sqlmesh.runner) to support granular check execution via registry.run_checks(..., checks=...) while maintaining 100% backward compatibility for container keys ("rules", "sqlmesh") and legacy check aliases.
• Added comprehensive unit tests in packages/tff-core/tests/test_registry.py covering registry lookups, resolution, fallback runners, and granular check execution across dbt, Dataform, and SQLMesh runners.
• Verified 100% diff test coverage (526 lines covered, 0 missing) enforced by .githooks/pre-push.

🤔 Why is this change needed?

Resolves issue #146 (Phase 3 of the architectural improvements):

• Eliminates duplicated, desynchronized rule and check metadata definitions across health.py, report.py, and runner modules.
• Enables granular execution of individual rules or checks via tff lint --checks no_missing_owner,ban_select_star on dbt, Dataform, and SQLMesh projects.
• Provides a clean foundation for Phase 4 (external plugins and community rules).

🔗 Related Issues (Optional)

Closes #146

✅ Next Steps / Checklist (Optional)

• All 283 unit tests passing
• Ruff linter & dependency audit passing
• 100% diff coverage achieved and verified by pre-push hook
• Documentation updated in README.md