Fix nohup quoting: write train command to script file first

The train_cmd contains single quotes (optimizer args) which broke
nohup bash -c '...'. Now writes to /tmp/train.sh via heredoc then
runs nohup on the script file.

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

src/content_engine/services/runpod_trainer.py

@@ -595,12 +595,12 @@ resolution = [{resolution}, {resolution}]
             pid_file = "/tmp/training.pid"
             exit_file = "/tmp/training.exit"
             await self._ssh_exec(ssh, f"rm -f {log_file} {exit_file}")
-            # Run in background with nohup, redirect output to log, save PID and exit code
-            detached_cmd = (
-                f"nohup bash -c '{train_cmd} > {log_file} 2>&1; echo $? > {exit_file}' &\n"
-                f"echo $! > {pid_file}"
-            )
-            await self._ssh_exec(ssh, detached_cmd, timeout=10)
+            # Write training command to a script file (avoids quoting issues with nohup)
+            script_file = "/tmp/train.sh"
+            # Escape the command for heredoc
+            await self._ssh_exec(ssh, f"cat > {script_file} << 'TRAINEOF'\n#!/bin/bash\n{train_cmd} > {log_file} 2>&1\necho $? > {exit_file}\nTRAINEOF")
+            await self._ssh_exec(ssh, f"chmod +x {script_file}")
+            await self._ssh_exec(ssh, f"nohup {script_file} &\necho $! > {pid_file}", timeout=10)
             await asyncio.sleep(2)
             pid = (await self._ssh_exec(ssh, f"cat {pid_file} 2>/dev/null")).strip()
             job._log(f"Training PID: {pid}")