feat(core): broaden .understandignore starter — Ruby RSpec/Minitest patterns

[thejesh23]

Summary

• EXACT_DIR_NAMES += spec — RSpec's canonical home directory (used by discourse, homebrew, gitlab, fastlane, rubocop, and every rspec-rails scaffold). This is the load-bearing addition: spec/ alone drives 46 pp of the 51% weighted total reduction below.
• New TEST_PATTERN_GROUPS entry Ruby with 6 patterns (RSpec *_spec.rb, Minitest *_test.rb / test_*.rb, harness bootstraps spec_helper.rb / test_helper.rb / rails_helper.rb).
• +3 tests in ignore-generator.test.ts; ordering invariant extended to place Ruby after Rust.
• build + lint clean; ignore-generator suite 48/48 pass, core suite 898 pass (2 pre-existing wasm module resolution failures on scala-extractor.test.ts / swift-extractor.test.ts reproduce unmodified on origin/main).

SKILL.md unchanged — Phase 0.5 delegates to generate-ignore.mjs since a0155c5, so no inline duplicate to keep in sync.

Why these patterns (and not others)

Ruby is closer to C++ in ecosystem shape than to Rust: aggressive test clustering under known dirs, plus a few naming conventions the starter had never been taught about. The spec/ gap was the load-bearing miss — with it added, every RSpec-based project (which is most modern Rails-adjacent Ruby) sees the discourse-scale reduction the moment it opts in.

Rejected as too project-specific or ambiguous: **/*_shared_examples.rb (RSpec-only, files usually live under spec/support/ which is already caught by spec/), **/factories.rb (FactoryBot — under spec/factories/, same), **/schema.rb (Rails-generated, but path is db/schema.rb — surfaces via .gitignore extraction rather than a starter suggestion).

Kept spec_helper.rb / test_helper.rb / rails_helper.rb as separate opt-in globs because engines and multi-app monorepos occasionally reference them via require_relative from outside the canonical spec/test trees.

Token impact (measured across 10 public Ruby repos)

Methodology: gh api .../git/trees?recursive=1, Ruby source = .rb .rake .gemspec, 1 MiB ≈ 262K tokens. Two hero projects reported below — the highest percentage win and the highest absolute win — same pattern as PR #582.

Project	Before (analyzed)	After (analyzed)	Reduction
rubocop/rubocop	10.38 MB / ~2.59M tok	3.38 MB / ~0.84M tok	−1.75M (67%)
discourse/discourse	36.84 MB / ~9.21M tok	14.71 MB / ~3.67M tok	−5.53M (60%)

• rubocop — highest %. RSpec-dominant repo where spec/ was the bulk of the untouched source tree.
• discourse — highest absolute saving. 3798 files under spec/ — the single largest un-caught test tree in the survey.

Weighted total across all 10 repos: −11.80M tok / 51% on 92 MB of Ruby source — the highest of any language group added so far. Full breakdown (for context, not repeated in the hero table): ruby/ruby 59% / −3.66M, rspec/rspec-rails 61%, fastlane/fastlane 42%, rails/rails 2% (already covered by existing test/ dir rule), Homebrew/brew 1%, jekyll/jekyll 0%, shopify/liquid 3%, basecamp/kamal 0%.

The spec/ dir rule alone accounts for 46 pp of the 51% weighted total; the file globs add another 5 pp by catching *_spec.rb in gem lib/ trees, Minitest files that leak outside test/ in Rails engines, and the helper bootstraps.

Test plan

• pnpm --filter @understand-anything/core build — clean
• pnpm --filter @understand-anything/core exec vitest run src/__tests__/ignore-generator.test.ts — 48/48 pass (3 new Ruby tests + the spec/ dir test)
• pnpm --filter @understand-anything/core test — 898 pass, 2 pre-existing wasm-module failures also present on origin/main (unrelated: scala-extractor, swift-extractor)
• pnpm lint — clean
• Manual preview of generated .understandignore on a fixture with spec/, test/, plus a lib/foo.rb + lib/foo_spec.rb — Ruby group emitted after Rust, all six patterns present, # spec/ in the detected-dirs section, all commented

Closes #596