## Description
- Also started storing the embedding size and version in the table. This
will come handy while comparing overall size when different compression
and serialization formats are used..
- Also, it can be used to smartly auto-download the embeddings or inform
the user about approximate size when they decide to enable indexing or
search on a particular client.
## Tests
✅ Verified that embedding fetch and store is working fine.
✅ Verified that embeddings/diff API is working fine.
- Move deployment related stuff into a separate folder
- Document the overall approach better
- Add an Nginx specific environment and service definition
- Remove the restart limiter
During recent DDoS attacks, these restart limits had to be manually
disabled. For now, disabling them in our reference copy too, will bring them
back if they are needed later.
The use case for this came up in Discord - someone is running a custom self
hosted instance where they're using MinIO as their "production S3", but it is
not local (so they don't want the clients to connect without SSL). So we need a
way to get MinIO to work, which needs path style URLs, without also disabling
the SSL.
Add a dotted arrow between museum and the hot object storage to indicate that
clients talk directly to the primary S3 storage.
This has confused some users who were trying to setup self-hosting, them not
realizing that the minio buckets need to be accessible outside the container.
The actual flow is - for both GET and PUT, museum conjures up pre-signed URLs by
talking to the S3 bucket, and vends these pre-signed URLs back to the
client. Thereafter though (unless routed via a Cloudflare worker, which is not
the case when self-hosting), the clients directly connect to this presigned URL.
Maybe too much information to convey with a single dotted arrow, but at least
it'll give a hint that there is a direct connection there somewhere.
