Repair wp/5.6 branch Actions: Replace wp-env's database check with a MariaDB health check - #597

What?

Ports #75046 ("wp-env: Fix MySQL startup race condition causing database connection errors"), using #83751's health check command:

• mysql and tests-mysql get a health check (one MARIADB_HEALTHCHECK constant) and MARIADB_AUTO_UPGRADE, so existing databases get the user healthcheck.sh needs.
• wordpress waits for mysql, and tests-wordpress for tests-mysql, with condition: service_healthy.
• wp-env start no longer runs wp db check, its 30-second retry loop or the extra 4-second sleep.
• CHANGELOG.md gets trunk's entry, extended for the ping fallback; the README note on wp db with MariaDB before 11.4 now says wp-env start isn't affected.

Why?

wp db check runs mariadb-check in the cli container. The current wordpress:cli image's client (MariaDB 11.8) requires TLS, and WP-CLI passes --no-defaults, so it can't be turned off. MariaDB before 11.4 offers no TLS by default, so the check fails with TLS/SSL error: SSL is required, but the server does not support it and wp-env start stops. This branch pins MariaDB 10.5 (#622), so every job that starts wp-env would fail. Trunk replaced the check with a health check that doesn't use the MySQL client.

How?

depends_on, start and clean follow WordPress#75046 hunk for hunk, adapted to this branch's commands/start.js, commands/clean.js and wordpress.js. start/clean still start only mysql directly; tests-mysql (added in #628) starts through depends_on.

The health check runs healthcheck.sh --connect --innodb_initialized only when the data directory has .my-healthcheck.cnf and the image has healthcheck.sh. Otherwise it pings over TCP on 127.0.0.1 with mariadb-admin (the only name in 11.0+) or mysqladmin (the only name before 10.4), with -p"" for the empty root password. WordPress#75046's unconditional healthcheck.sh fails on versions the MariaDB setting can select:

• 10.2 and 10.3 images ship the script without the healthcheck user.
• Images built before 2022, and old patch tags such as 10.5.8 and 10.6.4, have no script.
• .my-healthcheck.cnf stays in the data volume, so an older image on a volume a newer one wrote would run a missing command (exit 127).

TCP rather than the Unix socket, because the temporary server that initializes a new volume answers on the socket before the real server is listening.

Testing:

• wp-env: Fix MySQL startup race condition causing database connection errors WordPress/gutenberg#75046's two compose-builder tests and Env: Add an option to choose the MariaDB version WordPress/gutenberg#83751's health check tests (run through /bin/sh with stub healthcheck.sh, mariadb-admin and mysqladmin, skipped on Windows) are ported. npm run test-unit -- packages/env: 72 tests pass. (test/cli.js can't load locally because nodegit isn't built; unrelated.)
• The generated Docker Compose file validates with docker compose config. The database becomes healthy on mariadb:10.5 through healthcheck.sh, and on mariadb:10.3 through the ping fallback.

Use of AI Tools

This PR was created by Claude Code under my supervision. All code should be treated as AI-produced and not yet reviewed by a human until this PR is marked Ready for Review.