Refs: #14

Refs: #13

Refs: #12

Since the unsuffixed key is saved in database, a precommit event
fetching an edited attribute doesn't download the new value.

Hooks require uploaded files to be accessible and open

Instead of doing storage attribution in test setup, we implement a test
hook (in test/data/hook.py) as would be done in real application. As a
consequence, we change the way s3 requests are mocked (using moto).
Namely, we introduce a "_s3_client" class method in S3Storage that is
mocked in test setup so that mocking operates in the test hook.

In migration tests, instead of instantiating a new storage object, we
reuse the one bound to Image's data attribute so as to benefit from the
active mock.

Otherwise, reading the Binary object with .read() (not .getvalue())
returns nothing since the stream is positioned at the end.

Accordingly, use .read() method in test.

no need for complicated code for parallel file upload.

S3 object deletion can be disabled via a 's3-auto-delete' config option.

This implementation provides an S3Storage that can be used to store entities
attributes in an Amazon S3 storage.

It comes with an STKEY rql function to be able to retrieve the S3 object key
for the entity attribute (similar to BFSS's FSPATH rql function).

S3 connection configurations are attached to the S3Storage instance, like:

  s3_storage = S3Storage(s3_bucket)
  storages.set_attribute_storage(repo, 'File', 'data', s3_storage)