LEKTOR*: Continue implementations

Also, turn the binaries into lib+bin. We do this to try to begin their
compilation before the codegen finished for dependencies, specially for
zbus which takes a really long time.

We have a big refactor for the NKDB crate. We remove the queue from the
database. The queue should not be in the database. We place it in the
lektord crate. We also will place the search functionnalities out of the
nkdb or the lektord crate, and place it into a new one.

Also, update dependencies and run clippy.

Other developments:
- Update to amadeus
- Make the KId just a u64. This is just a simpler and better
  representation! The thing is Copy, we can use atomics and not bizarre
  mechanisms to factorize some Arc<str>, etc. We still can know when a
  kara was updated or the file changed, and we no longer need to
  factorize playlists. Note that the relation `RemoteKId <-> KId` is now
  a one-to-one relation.
- We also eepercutate the fact that KId is now copy in Amadeus. More
  things don't need clone anymore and the iterators now will return the
  KId and not a ref to it.
- Rationalize routes for lektord
- Update the returned data from kurisu API
- Change how to get, read or write the playlists and the queue.
- Add a separated epoch to playlists and the queue that is incremented
  each time we try to write it (controled by the methods.)
- Change the payload crates to reflect that
- Remove async_trait from where we can