potential way of implementing discussion #587

[mreineck]

This demonstrates how variables depending on problem dimensionality could be merged into std::arrays, and how it potentially simplifies the code.

This is not yet complete:

• I'm not happy with the variable names yet. For example, I renamed nf1, nf2, nf3 to nf123, since the obvious candidate nf was already used for storing the product nf1*nf2*nf3, and I didn't want to do big refactorings for this demonstrations.
• there are more good candidates in the interface of spreadinterp, which I haven't looked at much, mostly because there is a clash between finufft_core.h, where integer quantities are usually BIGINT,and spreadinterp.h, where they tend to be UBIGINT. While it is possible to convert individual integers on the fly when crossing the interface, the same does not work for std::arrays, so I tought I'd ask for opinions before taking this any further.

[ahbarnett]

THanks Martin, THis is a good idea, in the category "incremental simplifications that do not affect performance", right?
Personally, since index arithmetic can product <0 integers, I would switch everything to BIGINT in spreadinterp, if this allows. I'm sure we could bring this in, esp, if it's needed for other simplifications.