Opening this issue for the discussion and opinions on what numpy build to use for quaddtype testing and releases.

• With the upstream work going on in NumPy for flexible finfo support, we need it the source build to test those integrations here with quaddtype.
• @mattip 's contribution on adding same_value casting requires the nightly build for all different platforms testing.
• For releases maybe we should stick to the LTS version

This can go in different workflows or we can have 2 different version of releases for quaddtype (LTS and nightly). Please feel free to drop the opinion and suggestions