ci: Add Cirrus CI config for aarch64 build tests

[matthewfeickert]

Description

• Follow up PR to PR Test aarch64 wheels #166
• Add Cirrus CI config for aarch64 wheel tests.
• Exclude .cirrus.yml from MANIFEST.in.
• Remove QEMU based config for time.
• Add Cirrus CI badge to README.
  • 

Suggested squash and merge commit message

* Add Cirrus CI config for aarch64 wheel tests.
   - The jobs take a large amount of memory and so modification
     of the default resources is needed.
* Exclude .cirrus.yml from MANIFEST.in.
* Remove GitHub Actions QUEM based aarch64 test. As the Cirrus CI takes minutes
 instead of over an hour there is now not a need for this test to run in
 GitHub Actions.
* Add Cirrus CI build status badge for the main branch to README.

[matthewfeickert]

These jobs just take a lot of memory

[matthewfeickert]

This PR now takes aarch64 wheel building from the slowest (by far!) to one of the fastest to test. :)

[jmduarte]

@matthewfeickert is it possible to move the wheel building for aarch64 on new releases to Cirrus CI as well as part of this PR?

See: https://cibuildwheel.readthedocs.io/en/stable/setup/#cirrus-ci

Apparently, we can also build wheels for macOS ARM (M1/M2) directly on Cirrus CI.

[matthewfeickert]

is it possible to move the wheel building for aarch64 on new releases to Cirrus CI as well as part of this PR?

See: https://cibuildwheel.readthedocs.io/en/stable/setup/#cirrus-ci

We could do that, but as there might need to be some discussion on how this works functionally for uploading to PyPI I would do this in a separate follow up PR. I think this should be doable though with some coordination.

[lgray]

Thanks @matthewfeickert for setting up cirrus! It's really nice having free arm machines out there, and even fairly easy to use at that it seems. I might use this over in coffea to start covering that (much larger) stack.

[matthewfeickert]

Thanks @matthewfeickert for setting up cirrus! It's really nice having free arm machines out there, and even fairly easy to use at that it seems. I might use this over in coffea to start covering that (much larger) stack.

Happy to! Yeah, for people who have written GitLab CI pipelines before it is a pretty similar syntax (mostly just learning syntactic sugars). I was impressed with how they let you request additional resources (seemingly without much penalty in job start time).

• Container jobs start out with 2 CPUs and 4 GB of memory.
  • c.f. https://cirrus-ci.org/guide/linux/#linux-containers
• Container jobs max out at 16 CPUs, but a single task can't request more than 8 CPUs.
  • c.f. https://cirrus-ci.org/faq/#are-there-any-limits

[chrispap95]

Thanks! This looks great and is a much better long-term solution for arm64!