[3.1.4 Backport] Error from handleChangesResponse not handled correctly

When a replicator requests a rev that requires an on demand import from sync gateway, if that document fails to be imported for whatever reason, we don't handle the error in that situation. This has a knock on effect to CBL being stuck in busy state waiting for the rev messages it has requested. 

It looks like we get the error passed correctly up the stack to line 584 in blip_handler.go but as the fatal error callback is nil the error is ignored. We should be handling this error correctly here.

Repro steps for an example:

1. Add doc through sync gateway to db without import on  
2. Take down sync gateway to flush cache
3. Update doc in bucket to add illegal _id property (prevents sucessful on demand import)
4. Bring sync gateway back up
5. Start one shot pull from CBL
6. CBL hangs in busy state

I reproed on 3.1.2 but code looks same in 3.1.3 too should should be issue there too.