Move nvcc builds back from c++14 to c++17

[valassi]

Until today I had downgraded nvcc builds to c++14 (even if gccc builds use c++17), because I was using cuda 11.1 which does not support c++17.

I was using cuda 11.1 becuase of the performance regression (ggttgg slower by 20%) described in #282.

With the move to templated FFVs described in #328, a move back from cuda 11.1 to cuda 11.5 is possible (no performance regression, the code is actually faster!) and necessary (as otherwise the builds take too long).

I am therefore moving back to cuda 11.5. This supports c++17, so this move will allow reenabling c++17 builds in nvcc. I will do this AFTER merging #328.

The move from c++14 to c++17 will allow cleaning up the code in various places (eg in the MemoryAccessAmplitude variadic templates, and more).

[valassi]

I have even moved further from cuda11.5 to cuda11.6, released last week (#334)

[valassi]

This is completed in PR #354 that I am about to merge, can be closed