Optimized port communication for scalar types

[cmnrd]

In reactor-cpp, every value relayed via ports and stored in actions is wrapped in a value ptr. The value ptr itself is a wrapper around the smart pointers of the stdlib and extends it by control for mutability. However, a smartpointer adds a small (but still significant) overhead if simple data types such as scalar values are used. This PR provides two different template specializations that implement the value ptr interface. The old one (based on smart pointers) is only used for non-scalar types. For all scalar types, a new optimized specialization is added that simply stores the value directly instead of using a smartpointer.

On my laptop, this improved performance of the RadixSort benchmark by about 1.7x. Other benchmarks likely also benefit from the change, but maybe not as much.

[tanneberger]

I ran every test locally everything worked and the changes looked fine to me.