Define < and isless for Partials

[devmotion]

Fixes the incorrect comparison in #790 involving NaN. With this PR:

julia> xs =Dual(0.0,NaN)
Dual{Nothing}(0.0,NaN)
julia> xs <0false
julia> xs >0false
julia>isless(xs, 0)
false
julia>isless(0, xs)
true

[andreasnoack]

It was a bit of a treasure hunt to find out why this was necessary. Hadn't realized that

julia> [0.0] < [NaN]
true
julia> (0.0,) < (NaN,)
false

It reminds me of an objection I had against the definitions in operators.jl a long time ago. I don't understand how it is sound to define any fallbacks for Any.