Negative exponent

[gopa810]

I was comparing results of this with publicly available library for FFT algorithms (libfftw.so) and to have identical results, I had to adjust exponent on the line 19 in fft.go:

cmplx.Exp(complex(0, -2*float64(i)*math.Pi/float64(len(data))))

Exponent should be negative, so I added negative sign before 2.

Otherwise this algorithm is 5 times slower that FFT from library, and 3 times slower when memory allocations are optimized for consecutive input vectors with the same size.