MAINT: prevent universal2 thin upload

[tylerjereddy]

Fixes #155 (check logs to verify though)

• it turns out that the universal2 jobs in CI
  include generation of the thin arm64 wheel artifacts,
  which then get uploaded with the undesired 11_0 names,
  as per this log for 1.8.0rc1:
  https://dev.azure.com/numpy/27346c6a-2774-4eac-bf85-e068127c0ccc/_apis/build/builds/21654/logs/200

• since the logs tend to disappear for Azure, it says:

uploading wheelhouse/scipy-1.8.0rc1-cp39-cp39-macosx_11_0_arm64.whl wheelhouse/scipy-1.8.0rc1-cp39-cp39-macosx_12_0_universal2.whl

• the solution I've opted for here is to do a verbose deletion
  of the *arm64.whl file in universal2 workflows only, followed
  by a listing of files in wheelhouse path to increase confidence
  when examining logs, etc.

• note that since we can't really test the thin arm64 wheels in
  CI anyway, another viable option would be to rename the thin wheels
  in the universal2 jobs to use 12_0, upload both of those wheels,
  and delete the individual thin arm64 wheel build CI jobs; for now,
  I've opted for the solution that is more convenient for me--we may
  be able to save electricity the other way, but we are also perhaps
  interested in removing the universal2 builds some day, which might
  be more work if we purge out the isolated thin arm64 builds now..

cc @rgommers @isuruf

[tylerjereddy]

should be backported once working, and before 1.8.0rc2

[isuruf]

I'm not sure why there are two jobs osx-Py39-universal2, osx-Py39_arm64. Only the former is needed.

[tylerjereddy]

I'm not sure why there are two jobs osx-Py39-universal2, osx-Py39_arm64. Only the former is needed.

See discussion above--it is not like I didn't think about this, so I'm open to other solutions, although this may be the least work for me for now.

[tylerjereddy]

Closed in favor of gh-157.