pytest: test that we properly apply channel_update.

[rustyrussell]

Spoiler: we don't, if it's in a routehint!

See also: https://github.com/ElementsProject/lightning/issues/4781

Pay should probably retry in this case?

[cdecker]

Ok, debugging this further it took me a while, but it appears l3 is serving a stale channel_update with the old fees, which l1 happily stores, and then tries the exact same path again (being the only one available). The gossmap is refreshed just before computing any route, by calling get_gossmap which refreshes internally, and I made sure we are seeing the channel_update alright, just the fees don't match our expectation.

@rustyrussell: I'm not exactly up to speed on the channel_update caching logic we use to defer updates in some cases, could this be related to that? I'd have expected the setchannelfee command to synchronously create a new update and stash it with lightningd which can then serve it, but we seem to have a stale one in there.