Signed overflow in zz_divremi

[percontation]

In the line r += b; in zz_divremi (zz.c:475), r and b can have large values that cause signed overflow.

For example, I believe running the t-zz test triggers a call where r == -4156049824137537374 and b == -7351016435385864854. (As noticed by ubsan, as I was trying to test unrelated functions I was adding.)

This is definitely a theoretical issue because signed overflow is undefined behavior in C, but have not attempted to determine if it's actually an issue that causes zz_divremi to have wrong values. (And, if there are no wrong values, an appropriate fix is just to cast things around for that addition.)

[wbhart]

As b is the only one of the two that can have value LONG_MIN, and r and b must have different signs, I think the arithmetic is correct (in twos complement at least). If there are compilers that complain, we should fix that with some casts though.