Replace the current delay timeout mechanism with a spinlock.
Both mechanisms try to eliminate the possibility of an in-flight
async op accessing state that has been destroyed by stop().
But the spinlock is less arbitrary.
This fixes the bug where in-flight async ops that were triggered
by DevReattacher (such as livoxGen1_attachDeviceReq) that have
unconditional delays in them would fire late, after the state
var they rely upon had already been de-initialized.
We use an async bridge to ensure that those ops get executed in
the background and then resume execution after the delay.
hayodea