Path to batched and vectorized execution primitives in core postgres executor

This session is aimed at hackers who have either built batched or vectorized executors elsewhere, or have tried to graft OLAP storage and execution onto Postgres and run into the executor boundary. The goal is a concrete working conversation about the primitives themselves: where batches enter and exit the plan tree, how a batched interface coexists with the existing tuple path, what the TAM and expression-evaluation APIs need to expose, and which incremental steps are tractable in the next cycle or two.

Recent work on the pgsql-hackers 'Batching in executor' thread has started in this direction, pulling page-sized batches of HeapTuples into Scan nodes using some new TAM APIs as a first step.