Adds a profile-gated `mysql9` service to compose.yaml (port 3307,
mysql:9.4) so the existing 8.0 instance on 3306 can run alongside it.
The 9.x service uses the `mysql9` Compose profile to keep `docker
compose up -d` defaulting to just 8.0 — explicit opt-in via
`docker compose --profile mysql9 up -d mysql9`.

Makefile gains:
  - `test-mysql9` — runs the full integration suite against the 9.x
    instance by overriding MYSQL_PORT=3307 and MYSCHEMA_TEST_DSN. No
    other code paths change; the catalog reader, parser, and diff
    are exercised against 9.x as-is.
  - `clean-schema-mysql9` — DROP/CREATE myschema_test on 3307,
    mirroring `clean-schema` for 8.0.

AGENTS.md gets a short section documenting the workflow.

Verified locally: `make test-mysql9` passes the full integration
suite against mysql:9.4.

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

Adds a matrix dimension to the CI workflow so the integration suite
runs against both MySQL 8.0 (the existing baseline) and MySQL 9.4 on
each push / PR. Codecov upload only happens on the 8.0 leg to avoid
duplicate coverage reports.

For each matrix entry we:
  - bring up the target compose service (`mysql` for 8.0, `mysql9`
    for 9.4 — the latter via the `mysql9` profile, matching the
    Makefile target),
  - wait for the per-port mysqladmin ping,
  - run `make test` and `make test-scenario` with MYSQL_PORT and
    MYSCHEMA_TEST_DSN overridden via job-level env so the same
    Makefile invocation works for both versions.

test/scenario/helper.sh: the mysql client args now read MYSQL_HOST /
MYSQL_PORT / MYSQL_USER from the environment (with the existing 8.0
defaults preserved), so the scenario suite reaches the right port
when CI sets MYSQL_PORT=3307.

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

The matrix CI sets MYSQL_PORT=3307 in the job env so the scenario
suite hits the mysql9 container, but the Makefile's
`export MYSQL_PORT := 3306` was an unconditional assignment that
overrode the env value (Make assignments win over env unless `-e` is
passed). The result: the 9.4 leg's `make test-scenario` ran the
mysql client against 3306 and failed with "Can't connect" against
the 8.0 service that wasn't running on that leg.

Switch the five DB-connection vars (MYSQL_HOST/PORT/USER/PWD/DB) to
`?=` so the env value wins when set, and export them explicitly on a
trailing line. `make test` / `make test-mysql9` / `make test-scenario`
verified locally on both ports.

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

Three inline comments:

Makefile + AGENTS.md
  - test-mysql9 now overrides both MYSQL_PORT and MYSCHEMA_TEST_DSN
    explicitly. With only MYSQL_PORT overridden, a caller who already
    has MYSCHEMA_TEST_DSN set in their environment would silently hit
    the wrong port (the `?=` template at the top of the Makefile
    skips the recompute when MYSCHEMA_TEST_DSN is already set).
    Setting both keeps the target self-contained.
  - AGENTS.md doc updated to match: explicit on overriding both, and
    notes the rationale.
  - (Copilot #2, #3.)

PR description (no code change)
  - The PR body still claimed mysql9 was profile-gated; that profile
    was removed earlier in the branch and the body was stale.
    Updated via `gh pr edit` to match the current always-on
    behaviour. (Copilot #1.)

Verified: `make test-mysql9` passes against the 9.4 compose service
with no MYSCHEMA_TEST_DSN preset and with one preset to 3306 (the
explicit override now takes priority).

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

Copilot review: the readiness loop ran 60 iterations and exited via
the trailing `sleep 2` (which returns 0), so a MySQL that never
became reachable wouldn't fail the step — subsequent test runs would
crash with a connect error and waste CI minutes.

Use the standard `timeout 120 bash -c 'until <ping>; do sleep 2; done'`
idiom: if the server doesn't respond within 120 seconds the step
exits non-zero. (The compose-up loop is left as-is — it retries
just the `docker compose` invocation in case Docker itself is slow
to settle, and pulls/starts succeed on the first attempt in
practice.)

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