(1) are they interruptible (sending down a new work unit discards the one in progress immediately) to minimize stales? This is a headache with my BFL FPGAs.
Yes they are - sending down a new work unit discards the one in progress immediately
(2) If not interrupted, will they scan the entire noncerange or abort after hitting a share?
(3) Do they report found shares immediately or at the end of the work?
They will continue to scan the entire range, every nonce found is sent back immediately.
(4) Do they implement notification on exhausting the noncerange without finding a share?
No - it will simply time out and be idle - you will have to send a new job before it times out.