AD on GPU backend

The [v1.1.0](https://github.com/weymouth/WaterLily.jl/releases/tag/v1.1.0) release makes the solver fully differentiable using [ForwardDiff.jl](https://github.com/JuliaDiff/ForwardDiff.jl) on the CPU backend of [KernelAbstractions.jl](https://github.com/JuliaGPU/KernelAbstractions.jl), but is currently not working on the GPU backend.