Stop building manylinux1 wheels (eventually)

[Qalthos]

It is increasingly awkward to build manylinux1 wheels

• They are not supported by platforms other than x86_64 (and i686 which we do not build currently)
• The images we rely on to build them are deprecated
• Those same images do not have versions of Python > 3.9
• manylinux1 itself is based on CentOS 5 which reached EoL in 2017

We can and will continue to build them for now, but we will have to drop support for it at some point. This issue is for tracking that.

[webknjaz]

This is an accurate summary. When I first implemented packaging, manylinux1 what realistically the most relevant target that covered most of the environments in use. manylinux2010 tag wasn't available to the end-users (due to the fact that the tags are embedded with pip and it only recognizes the ones that are known to it), manylinux2014 wasn't available either, for the same reason. Over time I wanted to challenge myself to see how wide the version range could be in the matrix so I kept it, even reverted the build machinery borrowed from pyca/infra for it to keep working.
But yeah, today it's reasonable to drop it.

P.S. There's also musllinux now for which there's a tracking issue at #233. It's quite relevant as well, given that many folks like running Alpine containers. And the wider packaging/installing ecosystem is ready to use these.

[webknjaz]

I've made some cleanup attempts at #437. We can reopen and build on top of it once it's decided that it's time to drop manylinux{1,2010,2014}. Python 3.9.0+ bundles a pip version that understands PEP 600 tags so there's no need to build manylinux{1,2010,2014} for these, only for older Python.
Note that even after dropping the image building, it should remain possible to build the wheels in the previously built images that are already published.