I'm curious what the use case/customer demand for something like vectors to be transactional is? It strikes me as something fundamentally less about strong consistency than most use cases in a database. Don't get me wrong, if it were free to have transactional, why not, but I can only assume this comes with profound tradeoffs in a workload contention sense.
A good example is TiDB's implementation of vector indexes. TIDB is also a MySQL-compatible distributed transactional database, which separates the storage engine for vector indexes from the transactional store through an extra replica in its OLAP engine (via TiFlash), avoiding resource contention.
