⚠️ This PR adds a grace period to detect_crashed_workers() to prevent multi-dispatcher race conditions. It is a correctness fix for an internal dispatcher heuristic — not a change to any security boundary (isolation, auth, allowlists, or credential handling). It has been classified as public_pr per the upstream SECURITY.md policy which states in-process heuristics are not security boundaries.

What does this PR do?

Adds a 30-second grace period to detect_crashed_workers() to suppress
false-positive reclaims of freshly-spawned worker PIDs. Under
dispatch_in_gateway: true with two or more gateways, _pid_alive()
can transiently return False during the fork → /proc-visibility
window. The sibling dispatcher then reclaims the task and spawns a
duplicate worker. The existing 15-minute claim TTL still catches
genuinely-crashed workers; the grace only covers the launch window.

HERMES_KANBAN_CRASH_GRACE_SECONDS overrides the default. Value 0
restores the prior immediate-reclaim behaviour and is set on the
kanban_home test fixture so existing crash-detection tests continue
to pass.

Related Issue

This PR is a companion to merged PR #23442 (fix(kanban): extend stale claim instead of killing live worker), which addressed the same
multi-dispatcher race in release_stale_claims() and closed
#23025 (recurring stall-loop symptom).

Related context:

• recurring stall loop: kanban workers repeatedly reclaimed with stale_lock, zero output across respawns #23025 (closed) — recurring stall loop in multi-gateway
  deployments; root issue closed by fix(kanban): extend stale claim instead of killing live worker (salvage #23071) #23442. This PR closes the
  remaining detect_crashed_workers path.
• kanban dispatcher: macOS zombie detection is a no-op — _pid_alive returns True for defunct workers #20015 (closed) — _pid_alive false-negative behaviour on macOS
  that this PR is also resilient to (grace period covers any
  short-lived false negative, platform-independent).
• [Bug]: Kanban stale claim locks from dead workers have no auto-cleanup — tasks permanently stuck until manual intervention #22926 (open) — broader stale-claim auto-cleanup; out of scope
  for this PR.

No exact-match upstream issue exists for the detect_crashed_workers
launch-window race specifically. Open to filing one if maintainers
prefer.

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Changes Made

• hermes_cli/kanban_db.py:
  • Add DEFAULT_CRASH_GRACE_SECONDS = 30 constant.
  • Add _resolve_crash_grace_seconds() helper that reads
    HERMES_KANBAN_CRASH_GRACE_SECONDS with validation/fallback.
  • detect_crashed_workers() now selects started_at and skips the
    _pid_alive check when time.time() - started_at < grace.
• tests/hermes_cli/test_kanban_db.py: 3 regression tests
  (test_detect_crashed_workers_skips_freshly_claimed_tasks,
  test_detect_crashed_workers_grace_period_env_override,
  test_resolve_crash_grace_seconds_handles_bad_env).
• tests/hermes_cli/test_kanban_core_functionality.py: fixture patch
  HERMES_KANBAN_CRASH_GRACE_SECONDS=0 on kanban_home so existing
  crash-detection tests retain pre-fix semantics.
• scripts/release.py: add steveonjava@gmail.com → steveonjava to
  AUTHOR_MAP (required by .github/workflows/contributor-check.yml
  for first-time contributors).

Diff: 4 files, +115 / -1.

How to Test

# Grace-period regression tests (3 new)
scripts/run_tests.sh tests/hermes_cli/test_kanban_db.py -- -k "crash_grace or detect_crashed_workers_skips or detect_crashed_workers_grace or resolve_crash_grace" -q

# Existing crash-detection tests continue to pass with fixture patch
scripts/run_tests.sh tests/hermes_cli/test_kanban_core_functionality.py -- -q

# Optional: reproduce pre-fix behaviour by disabling the grace
HERMES_KANBAN_CRASH_GRACE_SECONDS=0 scripts/run_tests.sh tests/hermes_cli/test_kanban_db.py -- -q

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits
• I searched for existing PRs to make sure this isn't a duplicate
  (closest is fix(kanban): extend stale claim instead of killing live worker (salvage #23071) #23442; this is the companion fix to a different code
  path in the same race class)
• My PR contains only changes related to this fix/feature
• I've run the affected tests; the 3 new regression tests pass
• I've added tests for my changes
• I've tested on my platform
  • Flag for review: verified on Linux only.

Documentation & Housekeeping

• I've updated relevant documentation — or N/A
  • N/A — env-var-only override, no doc surface changes.
• I've updated cli-config.yaml.example if I added/changed config keys — or N/A
  • N/A — HERMES_KANBAN_CRASH_GRACE_SECONDS is intentionally env-only.
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — or N/A
  • N/A — internal heuristic only.
• I've considered cross-platform impact — or N/A
  • Yes — SQLite + stdlib time only. _pid_alive() already uses
    psutil.pid_exists().
• I've updated tool descriptions/schemas if I changed tool behavior — or N/A
  • N/A — no tool changes.