I opted to have this return Integers instead of floating point values. Most uses of this function are probably looking for integers and this avoids the inefficiency of converting an integer to floating point which the user immediately converts back to integer. One consequence of returning integers is that this function throws a DomainError for Inf or NaN inputs. The user would need to check for and handle these values anyway; this just encourages the check before calling sigexp.

julia> sigexp(d64"-1.25")
(-1, 125, -2)

julia> sigexp(d64"1")
(1, 1, 0)

julia> sigexp(d64"1.0")
(1, 10, -1)

Fixes #130
cc @quinnj