ci: test Zarr natively on Linux ARM64

[jdotc1]

Add one Python 3.12 minimal-dependency configuration on ubuntu-24.04-arm to the existing test matrix.

The purpose is to exercise Zarr's integration with NumPy, numcodecs, and its other dependencies on native Linux AArch64. Zarr itself is pure Python; this adds coverage of the existing array and codec paths in that environment.

The job reuses the existing just setup and just coverage steps. The minimal environment excludes optional dependency extras.

Validation at ea9aac4: the native ARM64 job succeeded with 12,340 passed, 1,214 skipped, and 4 xfailed.

[d-v-b]

architecture-specific issues in imaging workflows.

any motivating examples?

[d-v-b]

asking because this project is pure python, there's probably not much arch platform sensitivity here, vs our compiled deps

[jdotc1]

@d-v-b I don't have a Zarr-specific ARM64 failure or an imaging reproducer to cite; the imaging wording was too broad. I agree that the architecture sensitivity is mainly in the compiled dependencies.

The value of this job is exercising Zarr's existing array and codec tests with dependencies such as NumPy and numcodecs on native Linux AArch64. It adds one Python 3.12 minimal environment and uses the existing just coverage command. The new ARM64 job at ea9aac4 passed with 12,340 passed, 1,214 skipped, and 4 xfailed. I've updated the description to state that narrower scope.

I also added the release-note fragment in 1909641.