A small bug in integer-length, assuming I understand what that
function is meant to be doing.  This lets it correspond to the inum
case at least.

	* numbers.c (scm_integer_length): On negative bignums, adjust
	mpz_sizeinbase to account for it looking at absolute value where we
	want ones-complement.

	* tests/numbers.test (integer-length): Exercise some negatives, in
	particular "...11100..00".

In the tests here, it's only ...11100..00 which is affected by the
problem, the others are already right, I just added them for coverage
and confirmation of what integer-length does.