[ci] restore aarch64 linux builds (fixes #6509)

[jameslamb]

fixes #6509

follow-up to #6517

Proposes:

• restoring the QEMU-based aarch64 Linux builds
• moving scikit-learn import earlier, to hopefully load libgomp.so earlier
  • details: [ci] aarch64 python-package job failing: "cannot allocate memory in static TLS block" #6509 (comment)
• allowing scikit-build-core to set the wheel platform tag, instead of doing it manually with mv statements

Notes for Reviewers

What happened to the idea of running these builds in Docker on the macos-14 runner?

it isn't possible because running docker on those runners isn't possible (click me)

ref: #6509 (comment)

I was hoping to be able to use Docker on the M1 Mac runner to do aarch64 linux builds... those runners already have an arm64 CPU, so we'd be able to do builds and testing without emulation, which should be much faster.

Unfortunately, that did not work. As of this writing, that runner does not support nested virtualization, so it isn't possible to run docker on it. I tried with colima, the docker Homebrew cask, and podman.

Some references:

• https://stackoverflow.com/a/78202969/3986677
• Install Docker on Mac runners actions/runner#1456
• Add docker and docker-compose to macOS actions/runner-images#17

So for now, this PR just restores the QEMU-based builds (using emulation).

[StrikerRUS]

🚀
Awesome research in #6509!

[github-actions]

This pull request has been automatically locked since there has not been any recent activity since it was closed.
To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues including a reference to this.