[ENH] Allow use of unsupervised models through Scivision

[evangeline-corcoran]

It's worth considering if this is something we want to priortise. From discussions with potential users in the ecology/biology fields there is a general sentiment of wanting to move towards wider application of unsupervised/one-shot/few-shot approaches as their is often a huge quantity of data available in these domains, but the resources/expertise to create quality training datasets is often lacking. Also, depending on the biological question its often not clear what should be labelled in images.

Using unsupervised approached may also allow users greater flexibility / make it more feasible to construct a whole customised pipeline from data loading to results / visualisation through scivision

[ots22]

I think there is nothing preventing unsupervised models being loaded right now. Is this issue to add some? Or is there a particular feature in mind?