Prefer workers with cached actor images during scheduling

[han2ni3bal-pixel]

Fixes #276

Changes

• atelet periodically reports image digests held in its in-memory cache to ate-api
• ate-api stores cache information in Valkey, keyed by node
• During ResumeActor worker selection, prefer nodes where all required images are already cached; fall back to the existing random strategy when there is no cache hit
• Local snapshot node restrictions remain a hard constraint; image cache affinity is applied only after that

Testing

• Unit tests
• Manual verification on Multipass + k3s: resume counter actor; Cache miss/hit behaves as expected; Redis contains imageDigests

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[Julian Gutierrez Oschmann (juli4n)]

There have been discussions around soft-constraints for scheduling based on node properties (like this), this not only applies to OCI images but other artifacts like volume snapshots. I would recommend we discuss this in #276 and agree on a path forward before turning it into an implementation.

FYI Michelle Au (@msau42)