Implement data-specific cartesian index

[charleskawczynski]

This PR defines a "datalayout-specific"-CartesianIndex, which we define getindex/setindex for on DataLayouts, and these implementations do not swap indices, improving gpu memory access patterns.

Our fill! method exclusively uses this new indexing pattern, since there is only one datalayout per broadcast expression. This means that we can specialize on that datalayout's index.

Main:

julia> benchmarkfill!(device, data, 3, "VIJFH" );
Benchmarking ClimaCore fill! for VIJFH DataLayout
BenchmarkTools.Trial: 10000 samples with 1 evaluation.
 Range (min … max):  76.520 μs …  4.431 ms  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     77.929 μs              ┊ GC (median):    0.00%
 Time  (mean ± σ):   78.673 μs ± 44.074 μs  ┊ GC (mean ± σ):  0.00% ± 0.00%

              ▁▃▄▆▇█▇▅▆▃▂ ▁                                    
  ▁▁▁▁▁▂▃▃▄▄▆▇█████████████▇▆▆▅▄▄▃▃▂▂▂▂▂▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▃
  76.5 μs         Histogram: frequency by time          81 μs <

 Memory estimate: 560 bytes, allocs estimate: 18.
Benchmarking array fill! for VIJFH DataLayout
BenchmarkTools.Trial: 10000 samples with 1 evaluation.
 Range (min … max):  35.840 μs … 348.976 μs  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     37.200 μs               ┊ GC (median):    0.00%
 Time  (mean ± σ):   37.383 μs ±   3.625 μs  ┊ GC (mean ± σ):  0.00% ± 0.00%

                 ▄▆▅▆▇█▅▅▅▄▁                                    
  ▁▁▁▁▁▁▂▂▃▃▄▅▇▇██████████████▇▆▆▆▆▅▄▅▄▃▃▄▃▂▂▂▂▂▁▁▂▁▁▁▁▁▁▁▁▁▁▁ ▃
  35.8 μs         Histogram: frequency by time         39.6 μs <

 Memory estimate: 960 bytes, allocs estimate: 30.

julia> @test all(parent(data) .== 3)
Test Passed

This PR:

julia> benchmarkfill!(device, data, 3, "VIJFH" );
Benchmarking ClimaCore fill! for VIJFH DataLayout
BenchmarkTools.Trial: 10000 samples with 1 evaluation.
 Range (min … max):  44.090 μs … 340.207 μs  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     45.370 μs               ┊ GC (median):    0.00%
 Time  (mean ± σ):   45.454 μs ±   3.293 μs  ┊ GC (mean ± σ):  0.00% ± 0.00%

             ▁▁▂▅▅▆███▄▁                                        
  ▂▂▂▂▂▂▃▃▅▅▇████████████▆▄▄▃▃▃▂▂▂▂▂▂▂▂▂▂▂▂▂▂▁▁▂▂▂▂▂▂▂▁▂▂▂▂▂▂▂ ▄
  44.1 μs         Histogram: frequency by time         48.5 μs <

 Memory estimate: 560 bytes, allocs estimate: 18.
Benchmarking array fill! for VIJFH DataLayout
BenchmarkTools.Trial: 10000 samples with 1 evaluation.
 Range (min … max):  35.660 μs … 292.658 μs  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     36.990 μs               ┊ GC (median):    0.00%
 Time  (mean ± σ):   37.080 μs ±   3.062 μs  ┊ GC (mean ± σ):  0.00% ± 0.00%

            ▂▂▃▅▅▄▆▇█▆▆▅▂                                       
  ▁▁▁▁▂▃▃▅▆██████████████▇▆▅▄▃▂▂▂▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▃
  35.7 μs         Histogram: frequency by time         40.3 μs <

 Memory estimate: 960 bytes, allocs estimate: 30.

julia> @test all(parent(data) .== 3)
Test Passed

[charleskawczynski]

Superseded by #1902