fix(node-agent): handle bridge spawn ENOENT + add pid:host for process detection

Two bugs causing deltacast recorder 500s:

1. startDeltacastBridge() had no proc.on('error') handler — ENOENT when
   deltacast-bridge binary not in container PATH crashed the entire node-agent
   process (unhandled error event). Added error handler that logs and clears
   _dcBridge so the sidecar start continues.

2. node-agent container lacked pid:host — _dcBridgeProcessAlive() scans /proc
   but without host PID namespace it only sees container PIDs, so it always
   returned false and tried to re-spawn the bridge (hitting bug 1).
   pid:host lets the /proc scan see the host's deltacast-bridge systemd process.

🤖 Generated with Claude Code

docker-compose.worker.yml

@@ -43,6 +43,7 @@ services:
     build: ./services/node-agent
     restart: unless-stopped
     network_mode: host
+    pid: host
     environment:
       MAM_API_URL:   ${MAM_API_URL}
       NODE_TOKEN:    ${NODE_TOKEN:-}

services/node-agent/index.js

@@ -145,6 +145,11 @@ function startDeltacastBridge() {
     }
   });
 
+  proc.on('error', (err) => {
+    console.error(`[dc-bridge] spawn error: ${err.message} (binary=${DC_BRIDGE_BIN})`);
+    _dcBridge = null;
+  });
+
   proc.on('exit', (code, sig) => {
     console.error(`[dc-bridge] exited code=${code} signal=${sig}`);
     _dcBridge = null;