Wire OpenLoops knowledge feedback

[andrei-hasna]

Summary

• Add opt-in knowledgeFeedback target config and CLI flags for command, agent, and workflow loops.
• Emit failed/timed-out/abandoned loop outcomes to the Knowledge CLI with stable dedupe ids and bounded secret-scrubbed evidence.
• Read bounded Knowledge context before agent starts and append it as read-only prompt context.
• Document the feedback loop and add E2E coverage with fake Knowledge/Claude CLIs.

Validation

• bun test src/lib/knowledge-feedback.test.ts passed.
• bun test src/cli/index.test.ts passed: 148 tests.
• bun test src/lib/scheduler.test.ts src/lib/workflow-runner.test.ts src/lib/workflow-spec.test.ts src/cli/index.test.ts passed: 208 tests.
• bun run typecheck passed.
• bun run build passed.
• bun run test:boundary passed.
• todos redaction scan on the staged diff passed: no secret-like values detected.

Full-suite note

• bun test reached 536 pass / 4 fail. The only failures were src/daemon/install.test.ts attempts to write startup files under read-only /home/hasna/.config/systemd and /home/hasna/Library in this sandbox. Rerunning src/daemon/install.test.ts with HOME=/tmp/openloops-install-test-home passed: 5 tests.