Improve dft

[shibatch]

This is a combined patch for improving DFT.

• The butterfly kernels are all replaced.
• 128-point butterfly kernels are added.
• Improved transposer.
• Improved generator for execution plan.
• Added fftwtest. This tester compares the output from SleefDFT to that from fftw.
• Added roundtrip test. These tests are used when fftw is not available. It generates random input data, executes forward transform, and the backward transform. Then, it compares the results with the original data.
• A few functions in helperavx512f.h are replaced with better ones.
• Kernels that use streaming instructions are optional to be compiled in.