Using new Cartesian tools for copy!

[lindahua]

We should reimplement the copy! function for abstract arrays using @timholy's new Cartesian tools.

[timholy]

I'm fine with that, and I also understand if people have reservations about Cartesian invading base too deeply. Since it's my baby I probably don't have the best perspective on this, so I will defer to the wisdom of others.

[StefanKarpinski]

Paradoxically, I want Cartesian to invade base because I want it to go away. To explain that a bit further, what I'd like is for this Cartesian code to serve to help us understand exactly what we need from language features to replace all of it.

[timholy]

That's a very reasonable thought. However, the reality is probably that many of the remaining applications in abstractarray are "trivial," and so won't teach us much more. reducedim, broadcast, and getindex illustrate more of its power. I haven't thought yet about a possible extension for setindex!, that might be very informative.

I'm happy do all the conversions myself, as anything "easy" would not take me long. But if others want to tackle some of these as an exercise in learning more about how it all works---and therefore what we need to do to ditch Cartesian---I'm happy to let you have at it.

[StefanKarpinski]

No, feel free. You can probably shave another 500 lines off of Base ;-)

[JeffBezanson]

At this point we should only use it where there are performance problems (e.g. SubArray) and/or where the existing code is ugly anyway.

[StefanKarpinski]

Or where we can avoid using linear indexing and thereby make the algorithm more generic.

[timholy]

Implemented in #5671.