Breaking: better dimensional indexing

[rafaqz]

Major overhaul of indexing to remove bugs and better use all the tools we already have.

• AbstractBasicDimArray was added, and AbstractDimIndices and other objects like DimSlices are now in it along with all AbstractDimArray. This should give more consistent behavior.
• Tuple{<:Dimension} now works like CartesianIndex so DimIndices can work like CartesianIndices
• DimIndices can be used to index like CartesianIndices but of course it will work accross any dimension permutions
• DimKeys are now called DimSelectors (both still exported) because its an array of <:Dimension{<:Selector}
• DimSelectors works just like DimIndices. This means you can use it to interpolate, e.g. with a Near selector.
• AbstractArray{<:Dimension} and AbstractArray{<:DimTuple} work like AbstractArray{<:CartesianIndex} but again work on any permutation. They use mergedims so you still get a DimArray back.
• mergedims now accepts a tuple, and ust combines the dim names into a new dim name, if you want that. Indexing uses this to name dimensions after indexing with an AbstractArray.
• eachslice on AbstractDimStack no uses the DimSlices object rather than a Generator. The DimSlices object is useful for groupby as well, so we may as well have for eachslice too.

Breaking:

• DimIndices works for StepRange now and is generally much more like CartesianIndices. So it does not hold a LookupArray in its dimensions. Just ranges. This means selectors will not work on it, and should not - its a simple indexing object without lookups.
• LinearIndexing on stack makes sense in terms of dimensions - st[i] is now st[Dimindices(st)[i]]
• AbstractDimStack propagates for longer in indexing - if not all layers are scalars the scalar layers become zero dimensional fill(x) even in getindex.

[rafaqz]

@sethaxen is this close to being ready.

I don't expect a full review as its huge and patchy. But you should be aware of this as it will effect your types.

The two biggest changes for AbstractDimStack are:

• Indexing an AbstractDimStack always returns another AbstractDimStack unless all layers are scalars. Than mean getindex makes fill(x) layers sometimes. I think you suggested this should be the behaviour in the past!
• Linear indexing on AbstractDimStack is now correct (I hope), all indexing go via Dimension wrappers (currently indexing with Int will map that over all layers, which is fine for same sized stacks but wrong with mixed layers.)

[sethaxen]

Cool! I don't have time to review, but I ran all of my test suites using this PR, and I didn't encounter any new issues.

[rafaqz]

Thanks! Great they all pass. Mostly the things that broke were awful or buggy anyway and probably not used.

[codecov-commenter]

Codecov Report

Attention: 104 lines in your changes are missing coverage. Please review.

Comparison is base (3316875) 85.60% compared to head (a85ebf6) 84.21%.

Files	Patch %	Lines
src/dimindices.jl	80.00%	24 Missing :warning:
src/stack/indexing.jl	65.62%	22 Missing :warning:
src/array/indexing.jl	80.00%	14 Missing :warning:
src/Dimensions/primitives.jl	78.33%	13 Missing :warning:
src/LookupArrays/selector.jl	61.29%	12 Missing :warning:
src/stack/methods.jl	53.84%	6 Missing :warning:
src/array/array.jl	87.87%	4 Missing :warning:
src/array/methods.jl	25.00%	3 Missing :warning:
src/stack/stack.jl	85.00%	3 Missing :warning:
src/groupby.jl	60.00%	2 Missing :warning:
... and 1 more

Additional details and impacted files

```diff @@ Coverage Diff @@ ## main #615 +/- ## ========================================== - Coverage 85.60% 84.21% -1.39% ========================================== Files 45 45 Lines 3862 4055 +193 ========================================== + Hits 3306 3415 +109 - Misses 556 640 +84 ```

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.