Fix shrink_be_int bounds check

[chipshort]

noticed a problem with it after rebasing #1807 onto main

The bounds check did not account for the fact that the first bit of the rest of the bytes could be different from the sign extension bits. So, for example i8::MAX as i16 + 1 would be 0000000010000000 in binary. The first 8 bits are 0, but remaining 8 bits still overflow, as 10000000 is -128

[webmaster128]

Not sure if I deeply understand the change. Would need some pen and paper here. But the explanation and the tests look solid to me.