Support boolean datalayout indexing

[charleskawczynski]

For Julia arrays, users can do things like:

julia> A = rand(3,3)
3×3 Matrix{Float64}:
 0.628506  0.704686   0.205372
 0.390219  0.192054   0.118027
 0.27103   0.0604534  0.297599

julia> A[A .> 0.5] .= 1
2-element view(::Vector{Float64}, [1, 4]) with eltype Float64:
 1.0
 1.0

julia> A
3×3 Matrix{Float64}:
 1.0       1.0        0.205372
 0.390219  0.192054   0.118027
 0.27103   0.0604534  0.297599

It would be really convenient for users if we could somehow support this with DataLayouts, specifically for AbstractData{Bool} fields.

This may be a bit a bit of work to support getindex and setindex!, since setindex! will require some sort of lazy view, containing the indices, but I think it's possible in principle.

This might be a nice complement to the space masks.

[charleskawczynski]

I started sketching something, but I think we might want to draw inspiration from what Base is doing, and make sure we don't duplicate too much code since there will be different requirements on the gpu

@inline function Base.getindex(
    data::VIJFH,
    I::VIJFH{Bool},
)
    s = singleton(data)
    # fdim = field_dim(s) # get a single field to compute logical indices
    # Ipre = ntuple(i -> Colon(), Val(fdim - 1))
    # Ipost = ntuple(i -> Colon(), Val(ndims(data) - fdim))
    # V = view(parent(data), Ipre..., 1:1, Ipost...)
    # (inds, ) = Base.to_indices(V, (parent(I),)) # Base.LogicalIndex

    N = count(parent(I))
    v = similar(parent(data), eltype(data), N)
    i = 1
    for CI in CartesianIndices(data)
        v[i] = get_struct(
            parent(data),
            eltype(data),
            Val(field_dim(s)),
            CartesianIndex(to_data_specific(s, CI)),
        )
        i+=1
    end
    return v
end

[charleskawczynski]

With the ClimaCore syntax, we could write this as extrema(data[mask::IJHMask]), which will forward into a extrema(Base.getindex(parent(data), .~parent(mask.is_active))).

We can also then define Base.:~ on Boolean datalayouts to support extrema(.~data[mask::IJHMask]).