Initial chi-squared veto implementation

[alecgunny]

Generalizes some of the SNR calculation code to make this feasible and plays around with some scaling on this front. Think the transform itself is still missing some normalization constants here and there because the values are coming out enormous, but the general idea seems to work.

This metric should be close to 1 if the observed data contains the signal plus some gaussian noise, so the idea with testing is to construct such a signal/observation combination and test that the values come out as expected, but still not fully resolved how best to do this.

• Figure out normalization in the transform class
• Get rid of debugging print statements once things are solved
• Figure out how best to handle scaling in the SNR calculations, see all the TODOs in gw.py

[wbenoit26]

Is this different than doing integrand = template * other.conj() * scale / psd? Do you run into precision issues?

[alecgunny]

yeah that's exactly it, this is the most stable implementation I've found