Documentation

Drives the custom-backend store through the full cutover lifecycle (cassandra -> migration-to-postgresql -> postgresql). Every phase writes and reads, so all three interpreters (Cassandra, dual-write, Postgres) are exercised on both paths.

Rows created before the cutover (d1: pure Cassandra writes, d2: dual-write) must be served by the pure-Postgres interpreter afterwards — d1 proves the background worker backfilled a row that only ever existed in Cassandra. The re-put of d1 during the backfill proves the Postgres interpreter upserts (UPDATE-as-insert, like the Cassandra store).