feat: cross-node cache warming

[Ziinc]

This PR adds cross-node cache warming to Cachex caches.

• persistent node selection, to prevent imports from different nodes
• warm on cache startup in non-blocking way
• stream batches to CacheWarmer genserver, avoiding costly full table export.

Possible future extensions:

• add to health check, to prevent request hits before cache is fully warmed.
• Source.Supervisor boot should wait for cache warming to be complete.

[chasers]

Commenting for comms but lets wait on this one. Lots of edge cases to consider.

[hauleth]

IIRC Atom.to_string call is not needed, as it is default behaviour of comparing atoms.

[hauleth]

Sorting and then fetching first is the same as finding minimum.

[hauleth]

Suggested change

	|> Enum.filter(&(&1 != nil))
	|> Enum.sort_by(&Atom.to_string/1)
	|> List.first()
	|> Enum.filter(&(&1 != nil))
	|> Enum.min(&<=/2, fn -> nil end)

[Ziinc]

Closing this, as:

1. db upgrades as allowed us to spin up multiple nodes at concurrently now. This makes this PR obsolete as it is built on the assumption of sequential node startups.
2. a more scalable and safer way would be to broadcast db data fetching across the cluster, instead of sharing cached values.
3. caches should pre-warm themselves