Intervalize contents of Aff

[dpsanders]

This attempts to fix #14 in the most naive way: make everything stored inside the Aff type (the internal component that does the affine calculation) into an interval when the Aff object is first created.

The result of the example in #14 is as follows after this PR:

julia> Affine(-1..1) + Affine(1e23) + Affine(2020.0) - Affine(1e23)
affine=⟨[0, 1.67773e+07]; Interval{Float64}[[1, 1]]; [0, 0]⟩; range=[-1, 1.67773e+07]

Due to the precision of Float64 I don't think we can expect a better answer than this.
However, with BigFloat we get

julia> Affine(big(-1..1)) + Affine(big(1e23)) + Affine(big(2020.0)) - Affine(big(1e23))
affine=⟨[2020, 2020]₂₅₆; Interval{BigFloat}[[1, 1]₂₅₆]; [-0, 0]₂₅₆⟩; range=[2019, 2021]₂₅₆

which is correct!

(Note that there are some missing promotions to do Affine(-1..1) + Affine(big(1)).)

This PR surely does not give the fastest code, but at least it seems to be correct...

cc @mforets @RockerM4NHUN

[dpsanders]

Tests need updating.

[dpsanders]

The strange number in the above calculation is coming from

julia> nextfloat(1e23) - 1e23
1.6777216e7

julia> x = interval(1e23) + interval(2020)
Interval(1.0e23, 1.0000000000000001e23)

julia> x.hi == nextfloat(x.lo)
true

[dpsanders]

In other words,

julia> eps(1e23)
1.6777216e7