reduce stack sizes across the board

By default, each thread consumes 8MB of memory for the stack which can quickly
stack as machines have more cores and, especially during status, more threads
are started than there are cores. This overcommitting is by design, but
at least we should make sure that memory doesn't grow unnecessarily.

Especially iterators know the code they execute, hence these versions
should have a way to tune the stack size to reduce the peak memory footprint.