Replace plottertypes by Singleton

[jpthiele]

As proposed in #108 this changes dispatch.jl to use Singletons instead of using the set of nested ifs in plottertype with ismakie() and similar.

Since MakieType was heuristic based I had to introduce the alternative AbstractMakieType and change the Makie extension accordingly by adding where {MakieType <: AbstractMakieType} where needed.
Then all tests ran locally without problem.

Now, extending GridVisualize with another plotting backend can be done without needing to touch the main package.
Nevertheless, introducing a new shorthand through const AnotherPlotterPackageType = PlotterType{:AnotherPlotterPackage}
is certainly nice and quickly done.

@j-fu do you know of any downstream package that might explicitly use MakieType?

And if you are fine with this we should test the alternative to the example to remove the !private! functions ismakie and similar.

[jpthiele]

Had a quick search on GitHub and all that came up were forks of GridVisualize or completely unrelated projects when it comes to MakieType.

[j-fu]

I think formally this would be a breaking change which by semver increases the major version number and we very much should try to stick to this. I would put this on a list with other possibly breaking changes, which together would be substantial enough to be worth a najor release.

[j-fu]

Ok, had a closer look.

I was first surprised by the nameof approach (I wasn't really awareof this) , but this works, as with

MMM=Plots
nameof(MMM)

still would return :Plots.

[j-fu]

The other surprise is the use of Union types here. Indeed, they give the correct subtype relations. However isabstracttype(Union{Int32,Int64})==false, and therefore it might be better to call them UnionMakieType etc.

[j-fu]

For the formal semver consistency we could just set const MakieType = UnionMakieType and keep exporting this, with some deprecation information.

[j-fu]

I think the PR then also should replace the implementations of the ispyplot etc. functions. These use the real dangerous heuristics, and they are used sometimes
(at least in my teaching notebooks...).

[jpthiele]

@j-fu I have rewritten the isxyz functions to use plottertype(Plotter) <: XYZType which should hopefully work, but it would be great if you could try that in one of your notebooks.
Since that requires additional manual labor for adding new types and the implementation is practically the same I have deprecated them with the hint to use that check instead.

[jpthiele]

Also, in regards to #110 I have moved all the deprecated things to a dedicated deprecated.jl file in src, including their exports which should make the actual act of removing everything easier in a later breaking release.

[j-fu]

Ok I checked this out and test this in a couple of projects. So far, it works, and I think it should generally be ok.

However I think that we should not deprecate ismakie, ispyplot etc. We have implementations without heuristics now and we should avoid unnecessary breaking changes.

[jpthiele]

@j-fu isxyz have been undeprecated so I think it should be ready to merge

[j-fu]

Ok, LGTM. Thank you!

The CI problems seem to be due to some python problems on linux and should be addressed in a different PR.