perf(Propagator): Always commit in batches

[provokateurin]

I'm not sure this is a good idea or not, but I've seen that parents are updated multiple times within the same request.
This should help, as they will only get updated once, but it might also result in more memory consumption if many changes need to be propagated within a single request.
We could also go half way and automatically commit if there are too many parents, like the scanner logic already does. This would at least limit the memory consumption and should still perform better.

[provokateurin]

Tests are red, but I think they just need some manual calls to commitBatch().

[provokateurin]

I tried to fix a few of these issues, but it looks like it's not easily possible this way. Some sort of global propagator that can commit all batches in all propagators of all storages at once would be necessary. Depending on how that is done it might also be possible to gain even more performance by having a single query for propagation instead of one query per propagator.

[provokateurin]

A global batch propagator makes this a little better, but it doesn't solve all problems.

Some code relies on the assumption that changes are propagated right away and I don't feel comfortable with breaking that assumption, as it could result in really bad behavior changes.