poetry currently has two different versions of canonicalize_name() floating around, which is slightly absurd - and what's worse, they don't even agree.

Since python-poetry/poetry-core#328, poetry-core has an implementation that agrees with the implementation in packaging - but poetry has its own implementation that continues to disagree with that. Sometimes poetry uses one of these, and sometimes the other.

Let's treat the version from packaging as the canonical canonicalize_name(), and get rid of both of the re-implementations.

I don't know that this fixes - or breaks! - anything in particular; the unit tests don't detect a difference. But two differing implementations is surely an accident waiting to happen.

Aside from being one-true-implementation, a potential advantage of the packaging version is that it returns a NormalizedName: which is a newtype around string, rather than a plain string. That opens the door to future work in which the typechecker can verify that places which expect canonicalized names do indeed get them.

(Will follow up in poetry-core, but this one will need merging first since poetry sometimes uses the poetry-core version).