Play Framework

fix(test): absorb JPA cleanup lag in JpaLazyPoolTest on Windows runners

Netty can flush the HTTP response to the test client before the server-side
JPAPlugin.afterInvocation hook closes the EM and returns the HikariCP
connection. On Linux/macOS the lag is sub-millisecond; Windows runners hit
the OS scheduler quantum (~15ms) and the next test's baseline assertion
fires before the previous test's connections are returned to the pool.

Spin-wait for getActiveConnections() == 0 (5s budget) before sampling the
baseline and before each WatchedMax.start(). Sharper than a fixed sleep and
no-op on platforms where cleanup is already complete.

Surfaced as a flaky failure of pingNeverLeasesConnection on windows-latest
in CI for v1.13.18; passes on ubuntu-latest + macos-latest.

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