Reject zero exponent in mp_root_n

[MagicalTux]

mp_root_n(a, b, c) validates b < 0 but not b == 0. A zero exponent passes the input check and reaches ilog2 = ilog2 / b, a division by zero.

The early-exit branches (b > INT_MAX/2 and ilog2 < b) are both false when b == 0 and a has at least one bit, so control always reaches the divide.

Effect is platform dependent: SIGFPE (process abort / DoS) on x86/x86-64, silently returns a wrong result on architectures where integer divide-by-zero does not trap. It is undefined behavior on all platforms.

Fix: change the bound to b < 1 so a zero exponent returns MP_VAL.

Confirmed with UndefinedBehaviorSanitizer:

mp_root_n.c:60:19: runtime error: division by zero

Reproducer: mp_set_u32(&a, 123456789); mp_root_n(&a, 0, &c);

[czurnieden]

This and some other little bugs (e.g.: perfect powers get ignored) have been fixed in #599 which has not been merged (yet?). But nevertheless: thanks a lot!