Update MatrixAlgebraKit algorithm specification

[leburgel]

Updates the decomposition algorithm specification for the changes of MatrixAlgebraKit.jl v0.6.5.

For the SVD algorithm I set DefaultAlgorithm as the default, since that should just pick the best option regardless of the storage type. For the QR algorithm I couldn't do that because of QuantumKitHub/TensorKit.jl#405, so there I explicitly set it to HouseHolder(; positive = true).

[leburgel]

CUDA tests are failing, but I see the same failure on main so it doesn't seem to be caused by the changes here.

The issue seems to be that calling

left_orth!(t::TensorKit.TensorMap{ComplexF64, TensorKit.ComplexSpace, 2, 4, CUDA.CuArray{ComplexF64, 1, CUDA.DeviceMemory}}; trunc::MatrixAlgebraKit.TruncationByValue{Float64, Int64, typeof(abs)})

with only a trunc specified by no algorithm, is processed further down to

left_orth!(A::TensorKit.TensorMap{ComplexF64, TensorKit.ComplexSpace, 2, 4, CUDA.CuArray{ComplexF64, 1, CUDA.DeviceMemory}}, VC::Nothing, alg::MatrixAlgebraKit.LeftOrthAlgorithm{:svd, MatrixAlgebraKit.TruncatedAlgorithm{MatrixAlgebraKit.SafeDivideAndConquer{@NamedTuple{}}, MatrixAlgebraKit.TruncationByValue{Float64, Int64, typeof(abs)}}})

with an explicit SafeDivideAndConquer SVD algorithm. This then leads to

ArgumentError: MatrixAlgebraKit.CUSOLVER() does not provide gesdvd!

This seems like a problem with the algorithm selection not properly taking into account the storagetype in MatrixAlgebraKit.jl/TensorKit.jl? I can open an issue, but not really sure which one of the two is giving the problem.