RFC #218: Python vendoring

[gsnedders]

@Ms2ger suggested I file an RFC for this.

Rendered.

Implementation: web-platform-tests/wpt#49752.

[jcscottiii]

@WeizhongX I don't think this will impact Chromium CI but I thought I would double check with you.

[WeizhongX]

@WeizhongX I don't think this will impact Chromium CI but I thought I would double check with you.

We might need to create vpython wheel for the vendor packages, but should be fine for us.

[jgraham]

This generally seems fine to me, and looks like a significant improvement over the status quo. I have the following comments:

• I think the risk of this affecting Mozilla's prioritization of improving our story around non-vendored deps is minimal. Most of the pressure there really comes from deps with binary components, which aren't changed either way by this proposal.
• I think there is some risk that vendoring changes in the future in a way that breaks our use. It's specifically documented as being unstable and only designed to work with pip. I think it's a managable risk, but a real one.
• There is also the risk that this breaks unshared tests using wpt, if they are doing custom path manipulations. That isn't any kind of blocker, but people should be aware, since it may affect syncing.
• The implementation should be tested with full runs of wpt using the trigger branches, not just relying on CI jobs, since there's a risk of any test that does something custom with the Python path breaking.