Task: Enable multi-node dataset routing

[orlandohohmeier]

Summary

Update scheduler-to-worker dataset routing logic to support multiple data node targets per fetch.

• Allow a Fetch request to include multiple peer IDs (data nodes) with a random selection strategy.
• Adjust the scheduler to return a list of data nodes for each slice.
• Ensure fallback behavior is graceful: if one data node fails, retries will hit another peer.
• Reuse and extend the existing Reference::Peers variant and SelectionStrategy::Random.

Background

Currently, a worker fetches a slice from only one data node, which is a single point of failure. This change enables fault-tolerant fetching and prepares the system for distributed data delivery at scale.

[orlandohohmeier]

Adjust the scheduler to return a list of data nodes for each slice.

Returning a reference that is then resolved like all the others via the connection should suffice as that comes with the right strategies already, doesn't it?

[orlandohohmeier]

@nfnt is this something you can look into?

[nfnt]

Yes, I'll have a look.