Documentation elsewhere refers to this internal path, let's keep
it.

By using light gzip compression we can save a lot of disk IO during
the backup.

During the backup we create an intermediate copy of two directories:
builds and uploads. Instead of creating many small files with 'cp
-r', we now use tar (and fast gzip) to create single intermediate
files. This saves on disk IO and disk space while creating a backup.

This adds support for AWS S3 SSE with S3 managed keys, this means the
data is encrypted at rest and the encryption is handled transparently to
the end user as well as in the AWS Console. This is optional and not
required to make S3 uploads work.