Play Framework

fix(test): probe /countRo materialization via response body, not pool state

Two CI cycles on windows-latest both failed at the /countRo pool-active
assertion with observed max=0, despite the watcher sampling at 100kHz
(parkNanos 10us, sub-ms cross-platform). The lease+return cycle on a
warm pool is microseconds, and on Windows runners the 50-way concurrent
load resolves too quickly — or via too few client-side connections — to
guarantee any observable peak active count.

The pool-state probe was always indirect: the test goal is "prove the
readonly tx materialized the EM." A direct equivalent is to verify the
response body — "select count(n) from Note n" cannot return a value
without a JDBC connection, so a correct "count-ro:0" body proves both
materialization and a connection lease.

The /count assertion (cold pool) keeps the pool-active observation as
the false-positive guard for /ping. Only /countRo switches to body
probe, since it's the one that's been failing.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>