Bring back occurrence tracking #457

Occurrence tracking was never fully migrated from the AMI offline desktop app to the AMI platform. Currently every occurrence has a 1:1 relationship with a detection. Some design is needed to determine at what stage of the pipeline tracking runs and after which images are processed, since we now support multiple backends and parallel processing. It would be ideal to have a dedicated similarity model to generate a feature vector. Currently we are using the last layer output of the active species classifier.

For implementation approaches, refer the clustering algorithm currently in testing #774