While it fixes panics during careless use, it introduces a bug where an enormous number of samples gets dropped even when doing a single-threaded test:
This happens on several boxes.
I don't know what the bug is and I don't have time to investigate. Between having panics with careless use and bogus results I take the panics for the time being. Thus I propose a revert until someone can find out what's going on here.
I made sure that changes made past this are still there.