Asynchronous cache for zarr chunk loading

[fm3]

Switches the chunk cache inside zarr array implementation from a synchronous to asynchronous one. This means that concurrent incoming requests that are all cache misses do not each trigger chunk fetching. Only one does and the other requests hook into that.

I use the caching library from akka for this. Side effects include:

• The new cache is no longer LRU but LFU (frequency based). I do not think this will have much noticeable impact.
• The new cache has expiry times, which I think is a nice addition made simple by the new akka cache
• The signatures of many functions had to becomes asynchronous, which makes the code a bit less readable but should be acceptable. It also paves the way for other asynchronous action
• The exception handling inside ZarrBucketProvider could no longer use the tryo shortcut, instead I use Fox.recover to achieve the same effect.

URL of deployed dev instance (used for testing):

• https://zarralru.webknossos.xyz

Steps to test:

• Load remote zarr dataset, ideally one with very different chunk shape from the requested 32² (talk to me about a datasource-properties.json)
• The first few requests should cause a lot less load, be answered more quickly.
• wkw datasets and local zarr files should still load

Issues:

• contributes to Zarr backend follow-up: tech #6119

---

• Updated (unreleased) changelog
• Needs datastore update after deployment
• Ready for review

[jstriebel]

LGTM 👍