Make integral index type a compile time choice

[PDoakORNL]

With distG4 and the need to support linear indexing into large tensors int is not always going to cut it as a type. Make this a compile time choice i.e. template parameter, type alias etc. so this is the last time its an issue. Do this in such a way that we use the compiler to protect us from the possibility of a runtime overflow.

[gbalduzz]

why not just use std::size_t (uint64) for function's size? There should not be any performance penalty. We can not worry too much for integer loop variables inside kernels as no single dimension is ever larger than 2^31-1. Loops over the linear index can be problematic tough.