Easy, efficient, multiarch - can I pick all 3?

[shish]

I'm currently doing multi-arch builds with

          platforms: |
            linux/amd64
            linux/arm64

and this works, but the x86 build takes 45 seconds and the arm build takes 25 minutes, presumably due to being run inside qemu.

I feel like it should be possible to run the x86 build on an x86 runner, the arm build on an arm runner, and then combine the two results (in a third workflow that depends on the first two) -- but my google-fu is failing to find anybody who's done this before, let alone documented the process, let alone packaged the steps into some re-usable Actions. Has anybody done this / have any hints?

[crazy-max]

We have started to work on a reusable workflow to distribute builds across runners: #1321