Epic: clean up stale and broken agent/workflow runs (TTL GC, fail-fast on broken pods, console cleanup)

## Summary

Tracking thread for cleaning up **stale and broken agent/workflow runs** — both
the automatic controller-side lifecycle (TTL GC, fail-fast on broken pods,
resource cascade) and the interactive console-side cleanup (bulk delete, find
broken runs).

Today terminal runs accumulate forever, and runs whose pods are stuck on error
conditions (`ImagePullBackOff`, `CrashLoopBackOff`, unschedulable, config
errors) hang indefinitely with no expiry — leaving clutter in `kubectl`, the
console, and the cluster.

This is distinct from the existing *reporting* work (#171 surface failures to
status, #189 termination-log JSON, #187 SkillCard image verification), which
tells you *why* a run failed. This thread is about *reaping* what's left.

## Controller (konveyor/agentic-controller)

- [x] #198 — TTL-based garbage collection for terminal AgentRuns / AgentWorkflowRuns
- [x] #199 — Fail-fast and reap runs whose pods are stuck on error conditions
- [ ] #200 — Owner-reference cascade + orphan sweep on deletion

## Console (konveyor/tackle2-ui)

- [ ] konveyor/tackle2-ui#3556 — Bulk-delete / clear-completed for stale runs
- [ ] konveyor/tackle2-ui#3557 — Surface pod error conditions + filter for broken runs

## Related (not part of this thread)

- #171, #189, #187, #130, #166