Add --self-test flag to bin/validate-patterns (closes #71)

[JuanVqz]

Summary

Closes #71. Adds a --self-test flag to bin/validate-patterns that writes valid + invalid fixture YAMLs to a tmpdir and asserts the validator's behavior end-to-end. Wired into CI alongside the existing file-validation step. Single-file, stdlib-only, no Bundler / no test framework — matches the validator's existing philosophy.

Closes the test-coverage gap that was surfaced during PR #70 review: the validator's rejection paths (missing required keys, broken regex, unknown kind: value) are exercised manually but never asserted in CI. Without this, someone could refactor the validator and silently drop a guard while the 12 in-repo files keep passing.

What --self-test does

1. Builds a base valid pattern doc and 8 case variants.
2. Writes each case to Dir.mktmpdir.
3. Calls validate(path) in-process (not via subprocess).
4. Asserts:
   • Valid cases → errors.empty?
   • Invalid cases → errors contains a substring matching the expected rejection
5. Prints OK self-test (N/N passed) and exits 0; or prints which assertions failed and exits 1.

The 8 cases

#	Case	Expect
1	Valid: kind: breaking	pass
2	Valid: kind: deprecation	pass
3	Valid: kind: migration	pass
4	Valid: kind: optional	pass
5	Invalid: missing upgrade_findings: top-level key	fail with missing top-level key: upgrade_findings
6	Invalid: missing kind: per-pattern key	fail with missing key: kind
7	Invalid: pattern: regex [unclosed	fail with pattern regex does not compile
8	Invalid: kind: bogus	fail with invalid kind: "bogus"

Files changed

• bin/validate-patterns — added self_test function (~60 lines) and a --self-test dispatch branch. Updated usage docstring.
• .github/workflows/validate-patterns.yml — added ruby bin/validate-patterns --self-test step after the existing validation step.
• CLAUDE.md — documented the --self-test invocation under Repository tooling.
• rails-upgrade/CHANGELOG.md — Unreleased entry.

Test plan

• bin/validate-patterns --self-test exits 0 with OK self-test (8/8 passed)
• bin/validate-patterns (no flag) still validates all 12 in-repo files clean — no regression
• Manually verified one rejection case by introducing a bad fixture: validator's expected error string matches

Why this approach

The original issue body proposed a separate bin/validate-patterns-test script. Switched to a --self-test flag on the same binary because:

• One file, one entry point — no new binary to discover.
• Test logic lives next to the validator logic it tests.
• Calls validate(path) in-process — cleaner than shelling out and re-parsing exit codes.
• ~60 line addition rather than a 50-line standalone script.

Linked issues

• Closes Add --self-test flag to bin/validate-patterns (cover rejection paths) #71
• Refs Add kind: field to detection patterns; rename breaking_changes: → upgrade_findings: #53 (parent rollout — this is the test-coverage follow-up surfaced during the rollout)
• Refs Thread kind/bucket taxonomy through SKILL.md (follow-up to #69) #94 (SKILL.md threading follow-up — independent)

Out of scope

• Adopting RSpec / Minitest (the proposed flag deliberately uses no framework)
• Reorganizing bin/validate-patterns into a class
• Tests for behavior outside the rejection paths (e.g., file glob behavior, paths-empty warning)

[etagwerker]

@JuanVqz Looks good, thanks! 👍🏻