Ten thousand changes ought to be enough for any normal mapping activity.
Automatic edits and imports are controlled by scripts anyway so it doesn't
make a difference to them as they just have to adapt the limit.
The reason for my suggestion is that large changesets are becoming
increasingly difficult to handle due to frequent timeouts when trying
to load and process them.
The changeset size limit is returned by the API in the "capabilities" request
so client software that honours that information will automatically pick up
the new limit.