Conclusion:
Yes it is feasible to emit (wiki_db, revision_id) pairs. We got the prototype code to be able to emit these rows for a year worths of revisions for enwiki in about 5 mins of runtime.
Considering enwiki is one of the biggest wiki, this is very good news. We were also able to use Spark's JDBC Data source with parallelization set to 8 concurrent connections to the Analytics Replica. The replicas were able to handle the load quite fine.
Details on the shared notebook below.