fix(deploy): restart orphaned podcast worker + prevent start-limiter wedge (#115)

The bittorrented-podcast-worker systemd service is a separate long-running
process that `next start` does not recreate. When it crash-looped during the
June-7 mise/pnpm path drift it hit systemd's start limiter, went `failed`, and
stayed dead for ~2 weeks: podcast episode ingestion stopped on 2026-06-07
because deploy-droplet.yml only restarts the main app + iptv worker, never the
podcast worker, and its unit lacked StartLimitIntervalSec=0.

- deploy-droplet.yml: reset-failed + restart bittorrented-podcast-worker on
  every deploy (and reset-failed the iptv worker for parity); report its status.
- setup-server.sh: add StartLimitIntervalSec=0 to the podcast worker unit so a
  transient crash self-heals instead of wedging permanently.

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

Author: ralyodio

.github/workflows/deploy-droplet.yml

@@ -193,9 +193,18 @@ jobs:
             sudo systemctl restart bittorrented || echo "Warning: Could not restart main service"
             echo "✓ Main service restart attempted"
 
+            sudo systemctl reset-failed bittorrented-iptv-worker || true
             sudo systemctl restart bittorrented-iptv-worker || echo "Warning: Could not restart IPTV worker"
             echo "✓ IPTV worker restart attempted"
-            
+
+            # The podcast worker is a long-running process that is NOT recreated by
+            # `next start`; if it crash-loops (e.g. the mise/pnpm path drift) it hits
+            # systemd's start limiter and stays `failed` forever because nothing here
+            # restarts it. Reset the limiter and restart it on every deploy.
+            sudo systemctl reset-failed bittorrented-podcast-worker || true
+            sudo systemctl restart bittorrented-podcast-worker || echo "Warning: Could not restart podcast worker"
+            echo "✓ Podcast worker restart attempted"
+
             echo ""
             echo "=== Verifying deployment ==="
             # Give the unit a moment to either become active or crash so the
@@ -212,6 +221,7 @@ jobs:
               echo "failed" > "$STATUS_FILE"; exit 1
             fi
             systemctl is-active bittorrented-iptv-worker || echo "IPTV worker status unknown"
+            systemctl is-active bittorrented-podcast-worker || echo "Podcast worker status unknown"
 
             echo "Waiting for HTTP health check..."
             HEALTH_OK=false

scripts/setup-server.sh

@@ -933,6 +933,9 @@ WorkingDirectory=${DEPLOY_PATH}
 ExecStart=/bin/bash -c 'set -a; source ${DEPLOY_PATH}/.env; set +a; exec ${PNPM_HOME}/pnpm podcast-worker'
 Restart=on-failure
 RestartSec=30
+# Never let crash-looping wedge the worker in systemd's start limiter; it must
+# keep retrying so it self-heals after a transient failure (e.g. feed/network).
+StartLimitIntervalSec=0
 Environment=NODE_ENV=production
 Environment=PATH=${PNPM_HOME}:/usr/local/bin:/usr/bin:/bin