We're now able to rewind the instruction pointer in x86 backtraces. This
helps ensure addr2line cannot print information about unrelated adjacent
code. I've restored -fno-schedule-insns2 in most cases because it really
does cause unpredictable breakage for backtraces.