BigInt AddAssign<isize> with negative addend is off by 2^32 or 2^64

[HactarCE]

let mut x: BigInt = 0.into();
let y: isize = -1;
x += y;
assert_eq!(BigInt::from(-1), x); // fails
// x is actually 2^64-1

x + y gives the proper result (-1), and if y is i32 or i64 it works as expected, but isize somehow breaks things. When compiled for 64-bit the result is off by 2^64, and for 32-bit it is off by 2^32.

[cuviper]

Thank you for the fix in #133 -- I've also released this as num-bigint 0.2.6.

[HactarCE]

Thank you very much!