Addresses ppy#34705, I suppose.

The cagey tone of that statement is because this change merely papers
over the issue. The issue in question for the user that reported this is
that they have a bunch of very old beatmaps, whose md5 hashes do not
match the online hashes, that need updating. The submission/rank date
population was running every single time for these, and failing every
time, because there is really not much useful that the lookup *can* do.

Because mappers have made `OnlineID` essentially useless for determining
the provenance of a beatmap due to reusing them to "fix" beatmap
submission failures, online IDs have been explicitly disallowed from use
in any sort of beatmap lookup flow. The only things that are allowed to
be used are: md5 of the beatmap, and filename as a fallback for very old
beatmaps / beatmap packs.

If the user has local beatmaps with md5 not matching online, chances are
that any metadata lookups are likely to fail or return bogus data. At
that point my personal feeling is that backpopulation flows should leave
such beatmaps well alone and the user should just go update the beatmap
themselves.

I am aware that updating 124 individual beatmap sets would - in the
current state of things - would probably be a ridiculously onerous thing
to do, and that people have been asking multiple times for a facility to
update all local beatmaps at once, but that discussion is out of scope
at this stage.