Per irc conversation with yonik, it's not clear if updateLog works correctly in all situations if there is no _version_ field in the schema...
- it's certainly not tested
SOLR-3432makes it clear there are at least some problematic sitautions
For these reasons, updateLog initialization should fail fast with a clear error if there is no _version_ field in the schema.
(we can always relax this restriction later if people write the code/tests for it)