Support LU factorizations

[ChrisRackauckas]

MWE:

using CUDA
lu(cu(rand(2,2)))

ArgumentError: cannot take the CPU address of a CuArray{Float32, 2, CUDA.Mem.DeviceBuffer}
unsafe_convert(#unused#::Type{Ptr{Float32}}, x::CuArray{Float32, 2, CUDA.Mem.DeviceBuffer}) at array.jl:318
getrf!(A::CuArray{Float32, 2, CUDA.Mem.DeviceBuffer}) at lapack.jl:575
lu!(A::CuArray{Float32, 2, CUDA.Mem.DeviceBuffer}, ::RowMaximum; check::Bool) at lu.jl:81
(::LinearAlgebra.var"#lu!##kw")(::NamedTuple{(:check,), Tuple{Bool}}, ::typeof(lu!), A::CuArray{Float32, 2, CUDA.Mem.DeviceBuffer}, ::RowMaximum) at lu.jl:81
lu(A::CuArray{Float32, 2, CUDA.Mem.DeviceBuffer}, pivot::RowMaximum; check::Bool) at lu.jl:279
lu(A::CuArray{Float32, 2, CUDA.Mem.DeviceBuffer}, pivot::RowMaximum) at lu.jl:278
lu(A::CuArray{Float32, 2, CUDA.Mem.DeviceBuffer}) at lu.jl:278
top-level scope at test.jl:42
eval at boot.jl:373 [inlined]

It would be good to then make the ldiv! supported as well, similar to #1192

[maleadt]

Should be fixed by #1449