Add stalled-work and evidence-based debugging skills

[csark0812]

Summary

Add the i-need-help process skill to stop agents after two unproductive attempts on the same blocker, preserve the attempt history, and request specific help. Add evidence-based-debugging for reported failures and narrow probe to hunch and claim assessment. Update workflow routing, registries, documentation, and conformance coverage.

Validation

• npm run check — passed (123 tests passed, 5 skipped; audit and offline suite discovery passed).
• npm run agent:test:live -- i-need-help --workers=1 --retries=0 --timeout=600000 — 2/2 passed with captured skill-use and stopping traces.
• Companion live runs: 15/15 evidence-based debugging cases passed; 20/20 skill-enabled cases passed. One no-skill probe comparison failed and is retained as a baseline result.

The local full check required outer-sandbox escalation for Git-worktree fixture tests. The feature-specific live suite also required that execution context to capture Shell traces.