use eachindex and axes in loops

[heliosdrm]

Replace syntax for i in 1:length(x) by for i in eachindex(x), and for i in 1:size(x,d) by for i in axes(x,d), as recommended.

[Datseris]

As recommended by whom and why? :D I don't think it makes much of a difference but sure.

What I've noticed, that most likely will make a lot of difference, is that we do

    for j in eachindex(y), i in eachindex(x)
        @inbounds d[i,j] = evaluate(metric, x[i], y[j])
    end

instead of using any of the pairwise, rowwise, colwise from Distances.jl, which are much more performant versions that use BLAS and SIMD acceleration to compute distances across many pairs of vectors.

[heliosdrm]

The recommendation is here: https://julialang.org/blog/2016/02/iteration/

Also widely commented in this recent and very active post: https://discourse.julialang.org/t/discussion-on-why-i-no-longer-recommend-julia-by-yuri-vishnevsky/81151

That recommendation is done for the sake of generality; I don't think that the problems discussed there really apply in the code of this package. But I thought that if there is a consensus of for i in eachindex(x) being better Julia style than for i in 1:length(x), it makes no harm to do this little change and follow that style.

On the other hand, I don't understand this commentary:

What I've noticed, that most likely will make a lot of difference, is that we do

    for j in eachindex(y), i in eachindex(x)
        @inbounds d[i,j] = evaluate(metric, x[i], y[j])
    end

instead of using any of the pairwise, rowwise, colwise from Distances.jl, which are much more performant versions that use BLAS and SIMD acceleration to compute distances across many pairs of vectors.

No big change was proposed there, only replacing of ranges by eachindex. Do you mean that there are further changes that should be done?

[Datseris]

I didn't know that this was an established consensus. Now that I do, I will try to follow it. Sure, this is a straight forward stylistic improvement.

On the other hand, I don't understand this commentary

yeah, that was not really relevant with this PR, but I just saw it because it modified this source code. The short story is that Distances.jl recommends using pairwise when wanting to compute pairwise distances across collections of points. We don't do that here, and I believe because of that we miss some BLAS+SIMD acceleration that Distances.jl offers under the hood for some standard distances like Euclidean. It is a separate topic of course, so needs to be done in a different PR, but should definitely help with #130