Unify CUDA/HIP/OneAPI implementation

[marenz2569]

The implementation of the stress test for all three supported GPU families shares some functions and the general structure.
The OneAPI and Cuda class have the same arguments and the initGpus function has the same structure. Implementation specific code is already abstracted away in the create_load function.

DoD

• Unify high level interface of the stress test for all three GPU families
• Remove duplicated functions
• Use an enum class for specifying if float, double or automatic selection should be used