Reconsider warnings on multilayer leiden usage and low omega_end

[ragibson]

We currently warn on usage of multi-layer leiden since their package is not particularly optimized for that use case and is much slower than our unified louvain implementation tweaks. However, that implementation is long deprecated and my performance improvements to upstream leiden may have been enough to replace this warning with just a documentation note.

In contrast, there is a note in the documentation about omega_end being set too low in multi-layer optimization settings. This probably warrants a warning in the code itself if we detect it occurring (heuristically on the value of omega or explicitly on the returned omega estimates).