Add fine-tuning documentation

[johnbradley]

Provide documentation on how users could fine-tune BioCLIP.

[johnbradley]

@samuelstevens Suggested some example code from biobench:

binary SVM on vision models: https://github.com/samuelstevens/biobench/blob/main/biobench/newt/__init__.py
ridge classifier (kind of a linear probe): https://github.com/samuelstevens/biobench/blob/main/biobench/plantnet/__init__.py
simpleshot (sort of an upgraded KNN): https://github.com/samuelstevens/biobench/blob/main/biobench/birds525/__init__.py

[johnbradley]

My understanding of the three example above: These use only the visual part of the BioCLIP model and train SVM/ridge classifier/simpleshot based on the image embeddings. No text embeddings are used.

[samuelstevens]

Yep, that's correct. biobench is a method to evaluate vision-only models like the vision encoder from BioCLIP, or DINOv2, etc.