openclaw - 💡(How to fix) Fix feat: stop AI inference when source message is withdrawn/recalled

Problem

When a user sends a message and the bot starts AI inference, if the user withdraws the message during inference, the bot continues processing and eventually tries to reply. This wastes compute and may deliver content the user intended to hide.

Currently, the channel plugin can only detect the withdrawal after inference completes (when the reply send fails with Feishu error code 230011/231003). By that time, the full AI inference has already run.

Desired behavior

When the source message is withdrawn/recalled while AI inference is in progress:

1. The channel should be able to signal the core to abort the in-progress inference
2. The session should be cleaned up (no partial replies)
3. No reply should be sent

Use cases

• Private chat: User sends a message, realizes it was a mistake, withdraws it. The bot should stop processing immediately, not finish inference and then discard the reply.
• Group chat: User asks a question, bot starts processing, user withdraws. Depending on config, the bot may want to stop or continue — but the channel should at least have the ability to signal withdrawal.

Proposed approach

Add a channel event mechanism (e.g., onMessageWithdrawn(messageId)) that the core can listen for. When triggered:

1. Check if there's an active inference for that message
2. If so, abort it (similar to user-initiated cancellation)
3. Clean up the session state

This would be analogous to the existing cancellation mechanism but triggered by a channel event rather than user action.

Context

• Channel: Feishu (openclaw-lark)
• Feishu error codes: 230011 (recalled), 231003 (deleted)
• Related: larksuite/openclaw-lark#512 — plugin-level fallback for reply-on-withdrawn