Solving FFT instability issues

[zeeshansayyed]

This PR addresses #4

The error arises because of a check which in place to make sure that all the x_i values are real. This is done by the function check_xi_are_real() on line 261 of poibin.py. Due to instability in FFT, the imaginary parts don't turn out to be exactly 0. To ensure that these imaginary parts are indeed small, the following check has been placed in code:

xi_values.imag <= np.finfo(float).eps

On my machine, the value of eps turns out to be 2.220446049250313e-16

It turns out, this value is not enough to account for the instability in FFT. Increasing this slightly up to 1e-15 solves this problem.

I feel that making this change is a practical solution to the problem and will still not hurt accuracy.

[tsakim]

Thanks @zeeshansayyed, I'll try it out before merging.

[RoelantStegmann]

Any news on this? :)