This adds support for IndexLinear eachindex, as well as bounds-checking and getindex for a Broadcasted with an Integer index.

Instead of using the number of dimensions in eachindex to determine whether to use CartesianIndices, we may use the IndexStyle. This should not change anything currently, but this adds the possibility of nD broadcasting using linear indexing if the IndexStyle of the Broadcasted object is IndexLinear.

After this,

julia> bc = Broadcast.broadcasted(+, reshape(1:4, 2, 2), 1:2)
Base.Broadcast.Broadcasted(+, ([1 3; 2 4], 1:2))

julia> eachindex(bc)
CartesianIndices((2, 2))

julia> eachindex(IndexLinear(), bc)
Base.OneTo(4)

julia> [bc[I] for I in eachindex(IndexLinear(), bc)]
4-element Vector{Int64}:
 2
 4
 4
 6

julia> vec(collect(bc))
4-element Vector{Int64}:
 2
 4
 4
 6

This PR doesn't add true linear indexing support for IndexCartesian Broadcasted objects. In such cases, an Integer index is converted to a CartesianIndex before it is used in indexing.