Try S3 object storage for converted data

Following #821 (comment 24382)

Description

As a consequence we quickly explore using object storage:

• use object storage to store provider data, both source-data and json-data
• use the TSV representation for the series (we don't have to use JSON-Lines thanks to S3 that tolerates many small files better than file-systems)
• store a provider per bucket or a dataset per bucket, in function of Scaleway limits
• use S3 object versions
• update dbnomics-data-model and dbnomics-api to read from object storage instead of Git repositories

This would be a nice solution with many advantages over Git repositories:

• object storage is cheaper than block storage
• object storage is more secured against data loss
• revisions are implementable in a more efficient way than reading Git history
• more efficient that file-systems: less problems with many small files, or huge files
• no more problems with GitLab server (slow git clones, pushes...)

Tasks

• sync json-data after pipeline runs to S3 (@eraviart)
• adapt dbnomics-api and dbnomics-data-model to read from S3 (@cbenz )
• check that dbnomics-solr indexation script works (@cbenz)
• import Git repositories history for source-data and json-data as S3 versions

Before closing issue

• merge imf-fetcher!8 (merged)