fix: handle more bdb shutdown interrupts

[adam-miller]

This is an extension of the issue being addressed in #659

I encountered this during shutdown and it seems to just be an uncommon race condition where a thread happened to still be adding outlinks to the frontier. The added unit test reproduces the behavior if the Thread.interrupted() calls aren't present.

Maybe something could be done in ToeThread itself to override the interrupt() function and delay the base interrupt() call until it has reached the end of the chain?

[ato]

Yeah, to make it robust I guess we could give each thread its own "interrupt lock". Take the lock for the duration of each BDB operation and also when calling interrupt() on it.

[adam-miller]

If I understand Thread.interrupt() properly, it is just setting a flag, but that flag has special connotations for bdb. Maybe giving ToeThread its own requestStop() could replace the call to interrupt() and then switch uses of interrupted() to stopRequested(). Then maybe we wouldn't have to wrap each of the bdb interactions?