Move from #include magic to templates?

[mreineck]

I see that templates are being used more extensively in finufft development recently, and based on that I'd like to suggest to use templating also for the implementation of single- and double-precision NUFFT routines. Up to now, this is done via compiling some source files twice, with different macro definitions for FLT, CPX, etc. Switching to templates would most likely simplify the build machinery quite a bit, probably testing as well.
It would also open up a way to differentiate between the accuracy of the strength and position arrays, allowing, e.g., for 1D NUFFTS with single precision complex data, but double-precision locations of nonuniform points, which can be extremely handy in many circumstances. But that would of course (if at all) be a second step.
Switching to templates may look problematic when considering the finufft C bindings and the interface to FFTW, but this can be done witout much effort and I'd be happy to help with this.

[DiamonDinoia]

Yes, that is a good idea. I am already moving in that direction. At some point, we will switch completely.