add whittle processing scripts to track model performance

[kgarwoodsdzwa]

To assess tiny model performance, we need to compare device detections to real audio recordings over a certain time period, just until we determine a suitable confidence threshold to set for the detection -> alerting process. Scripts need to parse earthranger logged lora data for triggers and detections and pull out segments that correspond to the PAM audio to be checked.