Is there any particular reason for downcasting Float64 to Float32 but not Int64 to Int32? Should this be made more consistent? ```julia julia> Float64[1, 2] |> cu 2-element CuArray{Float32, 1, CUDA.Mem.DeviceBuffer}: 1.0 2.0 julia> Int64[1, 2] |> cu 2-element CuArray{Int64, 1, CUDA.Mem.DeviceBuffer}: 1 2 ``` Sorry if this has been already asked, it's hard to search for this specific issue.