Fix stop_on_error_timeout blocking other messages in queue

[glentakahashi]

In the current implementation, stop_on_error_timeout doesn't work as expected and instead completely blocks the message queue rather than operating as an async callback as intended.

Explanation:
stop_on_error_timeout = 0.1
If we send 100 requests at 1 per 0.01 second, and an error occurs after 0.3 seconds (request 30), this is what is expected to happen:

• requests 1-30 should be queued as normal
• requests 31-40 should be aborted
• requests 41-100 will be queued as normal

What actually happens:

• requests 1-30 are queued as normal
• the message queue blocks for 0.1 seconds
• requests 31-100 are queued as normal

A more extreme case can be seen if you bump this up to stop_on_error_timeout = 15. What happens in this case is:

• requests 1-30 are queued as normal
• the message queue blocks for 15 seconds! - No communication happens at this time, from the outside it appears as if the kernel is stalled.
• requests 31-100 are queued as normal

this PR fixes this by moving the "finish abort" callback to the event loop instead of using the existing message queue

[blink1073]

Thanks! This looks like a reasonable change to me, but I'll leave it open for a bit in case anyone else has any concerns.

[blink1073]

Thanks again!