fix(kanban): allow respawn after explicit unblock

[LeonSGP43]

Fixes #62418.

Summary

• ignore PR URL comments that were posted before the task's latest unblocked event
• keep guarding if a fresh PR URL comment appears after the unblock
• cover the resumed-review flow in the respawn-guard tests and dispatch integration

Verification

• /Users/leongong/Desktop/LeonProjects/worktrees/hermes-agent/.base/.venv/bin/python -m py_compile hermes_cli/kanban_db.py tests/hermes_cli/test_kanban_db.py
• /Users/leongong/Desktop/LeonProjects/worktrees/hermes-agent/.base/.venv/bin/python -m pytest tests/hermes_cli/test_kanban_db.py -k 'respawn_guard'
• git diff --check

[alt-glitch]

This was generated by AI during triage.

Duplicate of #46204 — both patch check_respawn_guard() in hermes_cli/kanban_db.py with the same mechanism: raise pr_cutoff to the latest unblocked task-event timestamp so PR-URL comments posted before the explicit unblock stop guarding respawn, while a fresh post-unblock PR comment still blocks. #46204 (2026-06-14) is the earlier/canonical fix. Both address issue #62418; the one-shot-override PR #62393 solves it via a distinct mechanism. Maintainer to pick.

[teknium1]

Thanks for the focused fix and dispatch-level regression coverage. The current-main premise is verified: check_respawn_guard() returns active_pr for recent PR comments without considering an unblock (hermes_cli/kanban_db.py:6881-6888), and dispatch_once() defers such tasks (hermes_cli/kanban_db.py:7249-7260).

Problems

• hermes_cli/kanban_db.py:6897 uses unblocked.created_at + 1 as the comment cutoff. Both comments and events use int(time.time()) timestamps (hermes_cli/kanban_db.py:2929, :3115), so a PR comment posted after unblock in the same second is ignored and can permit duplicate work. The new test sleeps past that boundary at tests/hermes_cli/test_kanban_db.py:1973, so it does not cover the ordering ambiguity.

Suggested changes

• Keep the cutoff inclusive at the unblock second and add a same-second regression test that expects active_pr; this conservatively preserves the duplicate-PR guard where ordering cannot be reconstructed from second-resolution timestamps.

Automated hermes-sweeper review.

[teknium1]

Both add_comment() and _append_event() persist int(time.time()) timestamps, so adding + 1 excludes a PR comment written after the unblock within the same second. That permits the duplicate-work case this guard is meant to prevent. Keep this cutoff inclusive at the unblock second and add a deterministic same-second regression test.