Shift from s3fs to boto3 for consistency

[metazool]

s3fs was picked arbitrarily, the object_store_api project uses boto3 and I agree with @albags that it's better to have consistency across projects (both for interfaces and for environment configuration)

Hopefully a small set of changes that are helpful to make while working through use of DVC for setting up a simple pipeline for extracting embeddings from images held in s3 (for building an initial classifier) #32

[metazool]

Closed in #35 (there's a way further to go with DVC, but this covered the object store python interface parts)