Test on macOS M1 where available

[hugovk]

GitHub Actions now has macOS M1 runners:

• https://github.blog/changelog/2024-01-30-github-actions-introducing-the-new-m1-macos-runner-available-to-open-source/
• https://github.blog/changelog/2024-01-30-github-actions-macos-14-sonoma-is-now-available/

Python 3.9.1 was the first to support Apple Silicon:

• https://www.python.org/downloads/release/python-391/

But 3.9 is not currently available for M1:

• Python3.9 on aarch64 for macOS 13.4.1. actions/setup-python#696 (comment)

So let's test 3.10-3.13 on M1, and 3.8-3.9 on Intel.

I bumped the PYTHONOPTIMIZE jobs from 3.8/3.9 to 3.10/3.11 to avoid complicating the matrix includes/excludes. And needed to install Ghostscript, it wasn't already on the image.

Here's how long it took to test on M1 (left) and Intel (right). M1 is faster for all, around twice as fast for 3.13 and PyPy3.10, and nearly three times as fast for PyPy3.9!

That's reduced from about 49 minutes to 25 minutes!

[hugovk]

GitHub Actions now has Python 3.9 available on both macos-13/Intel and macos-14/M1:

actions/setup-python#696 (comment)

• Python 3.9 on macos-13: 3m 44s https://github.com/hugovk/Pillow/actions/runs/10504390275/job/29101211666
• Python 3.9 on macos-14: 2m 22s https://github.com/hugovk/Pillow/actions/runs/10504390275/job/29101209375

That's 1m22s quicker on M1 :)

Shall we switch 3.9 from Intel to M1?

Or shall we keep at least one job on Intel to verify tests still pass on the older architecture? I favour this, it's not such a huge time saving and it's good to have more test coverage.