Collect SNR's of background events

[EthanMarx]

It would be great to have handy the SNR's of the background event's we collect, and save them in the Ledger. Not sure where the best place to do this calculation is. Maybe as a post processing step in the infer project? Once a segment is complete, construct a tensor of indices corresponding to the times of background events. Slice out these windows from the background into a tensor, and pass them through ml4gw.gw.compute_network_snr.

[EthanMarx]

Actually, would be great to have the individual IFO SNR's as well.

[EthanMarx]

At Erik's request, we should also calculate pearson statistics for events. I am doubtful this will add discriminatory power, but the more data the merrier. @alecgunny I know you have a gpu implementation living somewhere. Might be nice to add that to ml4gw for use here.

[alecgunny]

For background SNR, is this just the inner product of the background timeseries with itself (including PSD normalization)? Note that our existing implementation is intended for timeseries that don't include noise, but I guess I don't see why they should be any different. Worth noting that in ML4GW/ml4gw#79 I've made some changes to the SNR calculation code that could possibly help here, though maybe we already had everything we needed.

As for the pearson correlation, I've opened that up in ML4GW/ml4gw#78 if you want to take a look, that should be good to go. The cool thing about having these in torch is it makes possible supercharging the old vizapp to eventually having a GPU-accelerated real-time model analysis application that folks can play with.