run loop can busy-loop when the send queue blocks for a long period

[marten-seemann]

Apparently the Accelerated DHT client makes the send queue block for extended periods of time, at least on OSX: ipfs/kubo#9309.

This leads to busy-looping when the ACK-alarm expires, but we fail to send out the ACK packet.

[marten-seemann]

We probably need a proper re-design of our connection timer.

Requirements:

• the timer is always armed, there’s never a situation where it is not armed (at the least it’s armed for the idle timeout)
• there are situation when the timer fires, but no action can be taken
  • we might need to send an ACK, but the send queue is blocked, or we can’t send any more packets because we’re already tracking the maximum number of packets
• it might be armed again and again to the same (sentinel) value, for the pacing timer. This could be changed to arm to a different value (time.Now), at the potential cost of another call of time.Now