add pocketfft_cxx and duccfft benchees

[mreineck]

These two benchees complement the recent addition of pocketfft, since they have evolved from it.

pocketfft_cxx is a C++11 port of pocketfft with added multi-D transforms, vectorization and multithreading; it is currently used by scipy and jax and will be used by the next release of Eigen.

duccfft is a further evolution of pocketfft_cxx written in C++17 with many incremental improvements. The main goal was to make large, multi-D transforms faster (beyond, roughly, 1024x1024).

For compiling pocketfft, pocketfft_cxx and duccfft the following compiler flags (or their compiler-specific equivalents) should be used:

CFLAGS=-O3 -ffast-math -march=native
CXXFLAGS=-O3 -ffast-math -march=native -std=c++17

Accuracy should be very good (comparable with FFTW).
Performance is OK in general and good for large multi-D transforms.

NOTE: I'm currently disabling AX_CC_MAXOPT in configure.ac since I found no other way to pass my own flags to the benchees; see #4. This probably needs to be addressed before merging.

[stevengj]

Rebased to fix conflict.