Label Studio ML backend

[metazool]

This is the preferred option for #32 at the moment - a pre-annotation API for Label Studio

There's a draft on the label_studio_backend branch for this, presently with a single classifier with a hardcoded path, dropped into this repository for ease of testing.

What needs to happen with it is

• more testing against Label Studio, return labels it finds meaningful
• document the setup and link to docs for running Label Studio against the JASMIN store ( @Kzra )
• use a different model depending on s3 bucket
• (re)generate models with a lot more image input

There's a DVC pipeline that runs the last step. There are a lot more samples in the object store now, we need a means of tracking and searching what's been processed if it's not always going to run from the same place!