Could you clarify: which part seems to be working? The SKS servers certainly aren't, and the keyservers that are currently online don't appear to gossip or share keys with each other. That's why the post's dataset comes from querying the biggest/most popular ones manually.
> Was uploading a key somewhere a requirement for submitting to PyPi?
Where would "somewhere" be? If it was PyPI itself (or a server controlled by PyPI), replacing the key material would be trivial and would largely defeat the purpose of having signatures instead of just hashes.
In the past, "somewhere" could have been a gossiping SKS server. But that would tie PyPI's reliability and availability to that of the SKS network, which was never great even at its prime.
> Why were the keys not available from PyPi?
For the reason mentioned above: if PyPI is trusted to distribute the key material, then an attacker can simply replace the keys used to sign for the package. This makes it no better than having PyPI distribute hashes (which it already does), but a lot more complicated.
To my understanding, the reason PyPI originally accepted PGP keys is because someone asked for it and baseline expectations around security were more laissez-faire at the time: there was no baseline expectation that millions of people might be using `pip` (or `easy_install` at the time), and that all of them deserve the same integrity and authenticity properties as a small core of expert users. Those expectations have shifted over time towards the belief that ordinary users should also have signatures accessible to them, and I'm inclined to believe that's a good thing.