test(ci): add command validator regression coverage

[ayobamiseun]

Summary

validate-commands.js currently runs against the repository's eight valid command triples in CI. That proves the catalog passes, but not that the validator rejects the invalid states it exists to catch.

This PR adds a fixture-driven node:test suite that runs the real CLI inside temporary repositories and asserts its public exit codes and diagnostics. It then runs that suite in the existing command-validation CI job.

Coverage added

Six regression cases cover the validator's main contracts:

Case	Expected behavior
Matching command twins	Pass, including the plan to planning stem mapping
Claude command missing a TOML twin	Fail with the missing directory named
TOML command missing a Claude twin	Fail and count the canonical command correctly
Description drift across tools	Fail and print all three descriptions
Present file with no description	Fail with an actionable malformed-description diagnostic
Escaped quotes in TOML	Parse and compare the normalized description correctly

The passing stem-mapping fixture also uses single-quoted TOML, preserving coverage for that supported syntax.

Implementation

• Copies validate-commands.js into a fresh temporary repository for each test scenario.
• Writes only the command files needed for that scenario, then spawns the validator as a CLI.
• Removes every temporary repository after its test.
• Adds no dependencies and makes no production changes to the validator.

CI

The existing validate-commands job now runs:

node scripts/validate-commands.js
node --test scripts/validate-commands-test.js

This keeps the real-catalog gate and the validator self-tests together.

Verification

• node --test scripts/validate-commands-test.js - 6/6 pass
• node --test scripts/run-evals-test.js scripts/validate-commands-test.js - 18/18 pass
• node scripts/validate-commands.js - 8 commands, 0 errors
• node scripts/validate-skills.js - 24 skills, 0 errors, 0 warnings
• node scripts/run-evals.js - 124 checks passed, 0 errors, 0 warnings; rank-1 86%
• git diff --check upstream/main...HEAD - clean