CI: use Linux ARM runners.

[ncruces]

These just entered public preview:
https://github.blog/changelog/2025-01-16-linux-arm64-hosted-runners-now-available-for-free-in-public-repositories-public-preview/

[robherley]

@ncruces The new label is ubuntu-24.04-arm, I think there's a typo in the workflow definitions where you have *-arm64 and it's not matching any labels

[ncruces]

Nice catch. Just need to fix the DCO.

[anuraaga]

Shouldn't we add to the normal test task too?

Another note is personally, running this for every arch seems like overkill. amd64 was added for #385, while arm64 was here because it required qemu. Now it doesn't require qemu, so I'm wondering if we need both amd64 and arm64 in this matrix.

[mathetake]

yeah I would rather delete this for amd64 and arm64

[ncruces]

I used this because it's a “reduced” test suite compared to “full” (same way I used it for the VMs, although cross compilation helps there).

But I can drop this and run the full test suite on Linux ARM.

Or not at all (we stopped running any tests on macOS AMD64). I run a bunch of these other configurations (including BSD ARM) on my own repo, so it's not like it's untested.

[mathetake]

so up to you but we want to reduce the CI costs

[ncruces]

Well at least for this run, scratch+AMD was faster than base+AMD (and scratch+ARM is faster still).

I dunno what's cheaper by the minute?

If we wanna reduce costs we can drop some of the targets I added, or move them to something we manually run (e.g. before a release).

If we make them manually triggered I can at least run them for free on a fork, that's easy enough to do.