rfft -> fft and python3 command line usage updates

[evanmayer]

This example was super useful, but I had some trouble running it on the command line with python3.

Some changes:

• I believe the example constructed data is real, hence exploiting the more efficient rfft(). I ran it on complex data from an SDR, and changed it to fft() to handle this, and be more generically applicable.
• I also changed the power calculation slightly, but I believe it is equivalent.

You can take or leave these changes, of course! More importantly:

• A function was missing a return statement (which didn't seem to matter when running in the notebook, somehow).
• I explicitly used integer division where it makes sense to prevent promotion to a float that was raising errors in python3.

[telegraphic]

Hey @evanmayer -- fantastic stuff, happy to merge. Just a thought: would it be worth adding some example SDR data to the repo?

[evanmayer]

Sure! I'm sure several SDR users with gnuradio spectrometers have viewed this, trying to understand what the heck their gnuradio PFB block was doing, it could be relevant. What do you think the best way to get that in would be? I can stream some IQ data with a signal in it to a text file, which can be np.loadtxt'd in to the data var and processed, but curious if you had something else in mind.

I also had a poke and a prod with my SDR data and realized I omitted a change: x_summed = np.zeros((P, M * W - M + 1)) was throwing away the imag part of any data I passed in, giving incorrect results after the FFT. If you don't mind changing the assumption of real data to the assumption of complex data, I can remedy that in another PR with a dtype=np.complex128. I think the results should be unchanged for your real-only generated data.

Let me know! Thanks again for the tutorial and paper, amateur 21cm astronomy is grateful.