Improve aggregation speeds by using `eachindex` instead of `iterate`

[MilesCranmer]

This changes from e.g.,

  sum(loss(ŷ, y) for (ŷ, y) in zip(outputs, targets))

to

  sum(i -> loss(outputs[i], targets[i]), eachindex(outputs, targets))

which improves speeds by ~3.5x. Fixes https://github.com/JuliaML/LossFunctions.jl/issues/172