UDF use new schema migration

UDF is the "prototype candidate" for new, fine granular schema migration.

On the basic level, everything stays the same, more or less, except:

1. We might use separate versioning for migration messages (in addition to the messaging version)
2. For DROP X changes, we only encode what's necessary (the keyspace name, the table name, the timestamp). Not to save bytes, but to avoid going through a full schema merge on the node.
3. Include the type of change with all other schema changes, for the same reason.
4. For schema pulls, use a merkle tree or something similar, to only pull/apply the relevant changes.