> I also explained how an implementation may and may not use a fixnum check
> if the representation of a timestamp is not disjoint.

This comes in handy, but the current wording about the range of
timespecs is vague. IMHO we should either require a minimum range for
the seconds component (in bits) or require the implementation to
document its range (and to document whether or not it accepts a bignum
as the seconds component).