[roll] Roll fuchsia [kernel][irq] Fix IRQ re-bind edge cases.

What happens if an interrupt is bound to a port, the interrupt fires,
an interrupt packet is queued to the port, and before the packet is
read the interrupt is un-bound from the port?

Prior to this, the answer was "The packet is removed from the port,
destroying the interrupt's information, but the interrupt itself
remains in a state waiting to be acknowledged".

While this is quite the edge case (most folks don't do this) it is
rather confusing as we have apparently "lost" the interrupt.  Fix this
issue by effectively rolling the state of the interrupt object back.

When the interrupt is port-bound and has posted a packet, it is in the
NEEDSACK state.  If it becomes unbound, and the packet has not yet
been read, move back to the TRIGGERED state from the NEEDSACK state,
preserving the trigger time from the unread packet in the process.  If
someone binds the interrupt to a new port, a new packet with a new
bind-key will be generated with the original timestamp. Alternatively,
if someone waits on the interrupt object, their wait will not block,
but instead deliver the timestamp of the interrupt whose packet was
not originally read.  Either way, the interrupt will not be lost.

Test: new core tests
Original-Fixed: 492600836
Original-Reviewed-on: https://fuchsia-review.googlesource.com/c/fuchsia/+/1543085
Original-Revision: 15b94e583e4abb35ab969284e4554474b5013f49
GitOrigin-RevId: 566b3744ca30136992da01a5f1a054abaed754c3
Change-Id: I5591681c056d12ed1a61ba9b8f4234b8b1f2e849