Convert representation of samples/outputs from Vector{Tuple} to Matrix

[archermarx]

Following on from #338, when we transferred all of our sampling methods over to QuasiMonteCarlo.jl, we encountered the issue that QuasiMonteCarlo.jl outputs samples as a $d\times n$ Matrix, where $d$ is the dimension of the problem and $n$ is the number of samples, while Surrogates expects samples as a length-$n$ Vector of length-d Tuples. Converting Surrogates.jl to the former representation might improve speed and would allow us to remove the one remaining sampling-related bit of code, but would be a breaking change.