fix: reset idle timer on stderr activity, not just stdout

The idle timer was only resetting on stdout (user-facing results). The
agent-runner logs all activity — tool calls, task notifications, SDK messages
— to stderr. This meant the idle timer tracked 'time since last result sent'
rather than 'time since any agent activity', causing false stall timeouts
during legitimate long-running work like Task sub-agent research.

Now any stderr output resets the idle timer. The absolute timeout (45min)
remains as the hard ceiling. A genuine stall (frozen process, hung MCP server)
still produces no stderr and is correctly caught by the idle timeout.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: ziggy

src/container-runner.ts

@@ -365,6 +365,10 @@ export async function runContainerAgent(
 
     agentProcess.stderr.on('data', (data) => {
       const chunk = data.toString();
+      // Any stderr output means the agent is alive and working (tool calls,
+      // task notifications, internal logs). Reset idle timer so long-running
+      // sub-agent tasks don't trigger a false stall timeout.
+      resetIdleTimer();
       const lines = chunk.trim().split('\n');
       for (const line of lines) {
         if (line) logger.debug({ agent: group.folder }, line);