feat: direct s3 storage access for PGlite

[jgoux]

We are currently storing zipped tarballs of the PGDATA in S3 compatible storages. But it comes with complications:

• we must download and decompress the archive on the local filesystem
• we must have a cache strategy in place to clean the local filesystem

Connecting PGlite to S3 directly through s3fs is simpler as we don't need to deal with local filesystem and caching.

So far this PR works locally against Supabase Storage through the S3 compat layer.

We now need to test against a hosted Supabase Storage bucket.