Bayesian optimisation for multiple, parametrised tasks, includes GPU speedups using CUDA

Uses the Eigen linear algebra library, not included since large in size, please extract a release into the lib folder, then link in compilation as follows, (here using version 3.4.0)

g++ -I ../../lib/eigen-3.4.0/ eigenTest.cpp

Currently single task Bayesian optimisation is implemented, with a CPU only and GPU boosted implementation.

A basic CPU only Bayesian optimiser can be found at src/optimisers/Bayesian.cpp, with the GPU accelerated version src/optimisers/BayesianCUDA.cpp. The surrogate model CUDA kernel is located in src/optimisers/cuda/SurrogateKernel.cu - the GPU is used to increase the density of the Monte-Carlo surrogate model evaluations.

Polymorphism is implemented, primarily to simplify development so different optimisers and target merit functions can be combined, an example of this in action is seen in src/analysis/main.cpp.