Acknowledgement query validation

Plan: compare the existing bounded tuple-key DELETE with an exact-key temporary
MEMORY-table join before changing the query or raising the database floor.

MariaDB 10.3.39 and 10.11.19 both selected PRIMARY range access for the existing
1000-key DELETE on InnoDB and MEMORY, estimating 1000 matching rows in a 100000-row
queue. The join selected one PRIMARY lookup per scratch key. Five alternating
rounds per engine included clearing/refilling scratch keys in join timings.

These quiet local tests do not reproduce RTM concurrency or purge pressure, and
the servers use different default character sets. Do not compare their absolute
timings with each other. The 10.3 results disprove a blanket claim that this query
necessarily scans the entire queue below 10.6. No minimum-version change is made.

Run with mise-selected PHP and BOOST_DB_HOST/PORT/USER/PASSWORD pointing to a
disposable cacti_boost_contract database:
  php tests/performance/poller-ack.php

The script creates randomly named test tables, removes them in finally blocks,
and rejects any other database name. Raw plans and measurements are in the JSON.
