Is there documentation for implementing local storage archival and replay functions?

This would get tricky. The collector assumes workers have identical views.> ... that is available on all Worker Nodes.

Also, if you can, pointing Cribl to an on-prem Minio, can be the simplest solution

yep. Slightly more tricky, you could have a CIFS or NFS share that all the workers mount in the same place

but purely local would likely not work in a multi-worker scenario

NFS share for the win. Must be identical though. Very important.

https://docs.cribl.io/stream/destinations-fs/

https://docs.cribl.io/stream/collectors-filesystem/