19971226 djbfft 0.60, alpha.
19971226 performance: 783 788 815 825 for 256 points on a Pentium-100.
19971226 doc: wrote fftc4.3, fftc8.3, fftorder.3.
19971226 code: expanded accuracy.c to check both fftc4 and fftc8.
19971226 code: switched to dynamic allocation of speed arrays, to
         guarantee optimal alignment despite idiotic compilers.
19971226 code: expanded speed.c to time both fftc4 and fftc8.
19971225 code: various optimizations in fft.c.
19971225 code: took 32 out of fft.c.
19971225 code: split fftroots back into fftc8*.c, fftc4*.c.
19971225 code: added fftc4*. common 4/8 code in fft.c.
19971225 code: eliminated declarations of big transforms in fftc8.h.
19971225 code: split fftroots out of fftc8*.c.
19971225 code: merged fftc8_un.c into fftc8.c.
19971218 djbfft 0.55, alpha.
19971116 djbfft 0.50, alpha.
