Add controller API cancel_all_and_wait()

[Gin890]

cancel_all_commands() only starts the cancellation. It returns as soon as the cancel request is handed to the connection, and the host marks its command queue empty at that moment, so waiting for the queue proves nothing.

AbstractController::cancel_all_and_wait(timeout) cancels, then blocks until the device acks the cancel or the timeout passes. Per review, it uses the PABotBase2 reliable connection's wait_for_pending() instead of a no-op and a timer thread.

We didn't need to block on a cancel before because programs run serially and users can watch the result on the video capture screen. An AI agent needs the command and the in-game result to be accurately in sync to reduce its chance of making a mistake.

Changes

• ReliableStreamConnectionPushing: expose wait_for_pending(timeout). The PABotBase2 ReliableStreamConnection already implements it.
• CommandQueueManager::wait_for_cancel(timeout): if send_cancel() hit a full send buffer, the cancel is only stored and hasn't been sent yet. Keep retrying it until the timeout, then wait for the device to ack everything sent so far.
• AbstractController::cancel_all_and_wait() is now pure virtual:
  • PABotBase2 Switch controllers (Pro, wired, Joy-Con) and the keyboard: send the cancel under m_state_lock as before, then wait outside it.
  • sys-botbase: cqCancel has no ack, so it cancels and returns false.
  • Null controller: returns false.
• PybindSwitchProController::cancel_all_and_wait(timeout_millis) forwards to it. It returns false if the controller is not ready.

Semantics

• Returns true: the device has received the cancel. The firmware acks a packet once it lands in the receive buffer, before the command queue processes it. Messages are handled in order, so any command issued after this returns is queued behind the cancel, and the controller is usable right away.
• Returns false: on timeout, or when the controller can't confirm cancels.
• Throws: if the connection dies while waiting.
• Commands issued from other threads while this runs may land after the cancel and are not cancelled.

Testing

• The touched files compile with g++ and Qt 6 headers.
• CommandQueueManager::wait_for_cancel() was tested against a fake connection under ThreadSanitizer, with no races reported. The cases were:
  • normal ack
  • full buffer, then retry and send
  • timeout with no ack
  • timeout with the buffer stuck full
  • nothing queued
  • an earlier cancel not yet acked
  • the connection dying while waiting
  • a blocked wait_for_all() being released
• Not yet tested on hardware.

---

Generated by Claude Code

[Gin890]

Tested using Claude Cloud to rewrite the PR. It cost $6 using Claude Opus 5.5 Extra effort. Quite expensive! I can't imagine how much I had used in my day job by now.

I still need to read the rewritten PR to understand what the AI agent wrote

[Mysticial]

Approved, but rename the function from wait_for_all_commands_cancelled() to cancel_all_and_wait().