Re: ratio implementation

[Bill Schottstaedt]

If Guile wants to support "bignum floats" so to speak, I
think it could use gmp floats -- this strikes me as a big
task; I don't think its current dependence on doubles
and double->long conversions can be pushed very far.

I'll add a check in scm_rationalize for an error whose
inverse can't fit in a long and switch over to gmp in
that case, but I think the 0 error case should
use the simple version -- if someone wants pi to 10000000
decimals, he should ask for it explicitly.