Allow for geofencing of `speciesnet-classifier`

[nathanielrindlaub]

We currently only support geofencing when running SpeciesNet in speciesnet-all mode, which will run MegaDetector, the SpeciesNet classifier, and the SpeciesNet "ensemble" heuristic decision making logic (which includes geofencing) in a single step (see this issue in the animl-ml repo for more info).

However, we have seen in initial testing that there is still some utility in being able to call MegaDetector and SpeciesNet independently, for example if users want to adjust the default confidence thresholds on MegaDetector to make it more/less sensitive to animal detentions. We have found on Santa Cruz Island that running SpeciesNet in "all" mode misses a lot of rodent detections that MegaDetector will pick up on with our custom confidence threshold setting of 10%.

So if users are requesting MegaDetector and SpeciesNet Classifier predictions in separate Automation Rules, it would be nice to offer them the same geofencing functionality.

However, it may be challenging. SpeciesNet's geofence_base.json is 14 MB, making it difficult to load into the frontend and use in memory, for example if we wanted to apply the geofencing rules directly to the Automation Rules' Category Configs.

An alternative idea might be to apply the geofencing rules to the Category Configs on the API side, after users have saved the Automation Rule? This may require a bit more thought.