Range for the coeffients returned by gcdext

[Kevin Ryde]

Torbjorn Granlund <tege at swox.com> writes:
>
> Please don't specify something just for the uniformity of mpn.
> Uniformity isn't an important design goal of mpn.  Speed, though,
> is.

Oh, well, something nice for mpz_gcdext would be a good start.

I suspect people using both it and mpn_gcdext are probably relying on
getting something more or less like what we give now (the Euclidean
algorithm endpoint).