Rely on C89 <float.h> defining DBL_DIG

Rely on C89 <float.h> defining DBL_DIG

I would like to be able to assume that we have long doubles, and therefore
that LDBL_DIG and friends are all defined too. But it seems that we may
still support some platforms which are otherwise C89, but don't have even
trivial long-double support; in particular, HP/UX 10 apparently uses a
struct of four uint32_t values as long double, but doesn't support other
bits, and confuses the Configure probe that looks for quadmath.