GPU + truncation support collation issue

[kshyatt]

Here to track all the problems I find supporting truncation on CuTensorMap (and soon to be ROCTensorMap):

• findmin involves scalar indexing https://github.com/QuantumKitHub/TensorKit.jl/blob/main/src/factorizations/truncation.jl#L165. Needs a PR to GPUArrays.jl (in progress)