Profile compilation time

[bernhardmgruber]

I did some work on LLAMA, and a bit in picongpu, to optimize compilation times using clang's -ftime-trace. There were a few low hanging fruits which could speed up compilation a fair bit. For LLAMA, even by an order of magnitude for pathological data structure definitions.

I want to suggest to run such a trace on alpaka as well and just see whether there are any low hanging fruits as well.