Replace NEST typedefs with plain C++ datatypes

[pnbabu]

This PR fixes #2541. It replaces the following typedefs

• index, thread, port, rport with size_t
• weight with double
• delay with long

[JanVogelsang]

These constant values here are fixed at compile time, thus using constexpr is a good thing to do here.
Apart from that, everything looks right to me.

[JanVogelsang]

Edit: I didn't check yet if you missed any instances. Jochen, can you check that while you are reviewing?

[heplesser]

@pnbabu I have just merged #2616, which removes all the from-to asserts at the beginning of update() in models. This has unfortunately caused a lot of conflicts, but they should be straightforward to resolve by accepting the removal of the asserts.

[med-ayssar]

@pnbabu any updates?

[heplesser]

@pnbabu This looks good now overall, but having just read Stroustrup's "Tour", I am wondering about replacing so much with size_t (I know I argued for it before). The Core Guidelines don't say anything explicitly about size_t, but generally favour signed types (https://isocpp.github.io/CppCoreGuidelines/CppCoreGuidelines.html#Res-nonnegative) and point to suggest gsl::index for many cases where we currently use size_t, see https://isocpp.github.io/CppCoreGuidelines/CppCoreGuidelines.html#Res-subscripts. This leaves me wondering a bit. @jougs what to you think?

[JanVogelsang]

Looks good.
@heplesser Interesting, a few weeks ago I was asking myself the same question, as in university we got taught to use int for indexing, but personally I preferred size_t as well, because it seemed more descriptive. The core guidelines' arguments are valid, but I would have liked more depth, as I can't quite find anything regarding optimization/performance of unsigned vs signed types. I could only find an argument for unsigned types, as those can be divided by 2 faster than signed ones, but that's about it. And I am not sure why signed types enable better error detection either..
And apart from that, I don't really see an advantage of ptrdiff_t over size_t. In our code, we use size_t strictly for indexing or memory sizes, both of which can only be positive and will never be used in any operation that could return a negative value (or can you think of any?). And as we heavily rely on the STL, in my opinion, it makes more sense to stick with their "native" type.

[jougs]

Mostly fine by me. Just one repetitive pattern where I think the static_cast is not needed.