Fix overflows in mzd_init()

[tornaria]

The multiplication in mzd_init() causes a segfault in sagemath on 32 bit.

A doctest runs MatrixSpace(GF(2), 2^30)(1) expecting to get a memory allocation error, but it doesn't. In fact, the data member ends up with a malloc(0) (not NULL!) which later causes a segfault. All of this because (2^30) * (2^30/64) is 0 when size_t is 32 bits.

While looking at this I also noticed that mzd_init(1, INT_MAX) aborts (on any architecture), due to another overflow which I also fix here.

[malb]

Ta!