SNR calculation + tests

[alecgunny]

There's a couple things I think that are worth looking at in our SNR calculation

• Would it be more clear/consistent to refer to it as the "optimal ifo SNR", to avoid confusion with the SNR of a template waveform and a given signal (which would be normalized by this SNR)?
• Should we do some windowing before the FFT to get better behaved results? Should this window be an optional argument to the function?
• What is the appropriate normalization to do? I believe we want the FFT to have units of strain / Hz to make the SNR unitless, but is just dividing by the sample rate (as we do) correct? Do we adopt the scaling factor of sqrt(2 / N) used by the welch transform for the ASD? It's not clear to me why that has the correct units.
• What's the appropriate library to use as our gold standard for testing? Right now I can't really think of any tests a priori that don't just reimplement the logic of the function itself. Solved by fix SNR calculation in compute_snr_ifo, add test against lalsimulation #21

[EthanMarx]

I came across some lalsimulation utilities that perform SNR calculations in both frequency and time domain.

[alecgunny]

It actually looks like @deepchatterjeeligo already implemented those utilities in the tests. I'll edit the original message to be a checklist and cross that off, because the other questions are maybe still worth addressing