Issue: VS Code Extension: /auto command is ignored when sent while the agent is processing

Description
In the VS Code extension, when an LLM agent is actively processing a request, it is currently impossible to effectively queue the /auto command.

Steps to reproduce

1. Start a task that keeps the LLM agent busy (processing/thinking).
2. While the agent is still running, open the chat.
3. Type /auto and press Enter.
4. The chat enters a queued state because the agent is still processing.
5. Force-send the /auto message.

Current behavior
The /auto command is ignored. After the current processing finishes, the agent does not switch to Auto mode. The only workaround is to:

• wait until processing completes, or
• manually stop the current execution,

and only then send /auto again before asking the agent to continue.

Expected behavior
The /auto command should be preserved and executed once the current processing finishes. It should behave as a queued state change rather than being discarded.

This would allow users to enable Auto mode during long-running tasks without having to interrupt the agent or resend the command later.

Notes
The issue appears to be specific to commands such as /auto that modify the agent state while another request is still in progress. The command should remain pending and be applied as soon as the current execution ends.

Start by tracing the VS Code extension's handling of the /auto command, queued messages, and the agent's processing-completion path. Reproduce the reported sequence with a long-running task, then verify that a force-sent /auto remains pending and switches the agent to Auto mode when processing finishes.