ConstantInterpolation Extrapolation

[SouthEndMusic]

Fixes #378

I started working on this but then I found out this was already fxed by @DaniGlez in #379, but the issue wasn't closed. I did some nitpicking taking into account the dir of the ConstantInterpolation and the edgecase that there is only 1 datapoint.

On a more general note, I see that we use first and last a lot (also in this PR) which only works properly if u is a vector (or at least indexed as such). We also have specialized methods for vector versus matrix u. To ensure consistency and avoid code duplication, should we define custom indexing functions based on the type of u?

[ChrisRackauckas]

Add a test for this case?

[DaniGlez]

Yeah, it probably makes sense to close the issue, I forgot to use the magic keywords in the PR. Regarding this PR, the only issue I see is that this is consistent with the first / last open interval, but not with evaluation at the edge point, right? Currently, evaluation of a ConstantInterpolation at an edge point always returns the edge value regardless of extrapolation and dir settings.

julia> using DataInterpolations
julia> u = [2, 3];
julia> t = [0.0, 1.0];

julia> ConstantInterpolation(u, t; dir=:left)(t)
2-element Vector{Float64}:
 2.0
 3.0

julia> ConstantInterpolation(u, t; dir=:right)(t)
2-element Vector{Float64}:
 2.0
 3.0

Incidentally I just realized there seems to be another eltype bug with vector inputs:

julia> ConstantInterpolation(u, t; dir=:right)(t)
2-element Vector{Float64}:
 2.0
 3.0

julia> ConstantInterpolation(u, t; dir=:right).(t)
2-element Vector{Int64}:
 2
 3

[SouthEndMusic]

@DaniGlez Thanks for pointing out that boundary evaluation behaviour, I made a wrong assumption about it. Then this PR doesn't make sense.