From now on, test_download will run these tests. That means we benefit not only from the networking setup in there, but also from the other tests (for example test_all_urls to find problems with _VALID_URLs).
This moves playlist tests where they belong, i.e. to the extractors themselves.
Additionally, all our network interaction configuration for tests in test_download now applies to playlist tests as well.
If the result is not testable (for example, because a description changes often), either pass in a type or a regular expression (a string starting with 're:')
This is suboptimal, but at least this way we will need to look at the logs
only to check for network errors that happen too often, instead of
parsing a ton of lines each time to see if there is some true test failing