Quick visualisation of classification results

[metazool]

• Fully switches to the sqlite backend for embeddings + metadata
• Adds a classification column to the table, read into it from the API response to an image URL
• Add an optional normalisation function to the image preprocessing, not yet being passed through from the API
• Small improvements to documentation and tests

As ever this is work-in-progress but wanted to get it pushed through in order to switch tracks and look at our deployment options for pipelines - #47 - and how we can evolve this sensibly from a single VM deployment to (eventually) a managed k8s one with the option of a different workflow engine

The instructions for testing and running this should be covered in the README, probably some missing, also aiming to write some walkthrough documentation while watching pipelines run