[3.0][Testing] Add an integration test suite that runs against a real forum

[albertlast]

Description

The unit suite from #9326 is deliberately database-free, and its own bootstrap
says what is missing: anything reaching Config::$modSettings, User::$me or
Db::$db "belongs in an integration suite running against a real install".
There was not one, so most of the forum had no automated proof of anything.

This adds it. tests/Integration/ is a second PHPUnit testsuite that runs
against a forum installed by #9344, on either engine:

.docker/test.sh                     # both engines
.docker/test.sh --engine postgresql

composer test still runs everything. When there is no forum to talk to the
integration tests skip rather than fail, so it stays useful on a machine with
no Docker — 108 unit tests pass, 20 integration tests skip, exit 0.

What the base class provides

IntegrationTestCase gives each test:

• a transaction, rolled back in tearDown(), so tests need not order themselves
  around each other. $modSettings is restored separately, because the rollback
  returns the table and not the static array;
• actingAs($id) / adminId(), via User::setMe() — the same seam
  Login2::DoLogin() uses once it has checked the password;
• hook($name, $function) with permanent: false, so it lives in $modSettings
  and disappears with the rollback;
• assertNoErrorsLogged(), usually the point of the test. SMF records most of
  what goes wrong in log_errors rather than showing it, so a page that returned
  the right thing while quietly logging an undefined index has still regressed;
• queryRow() / rawSetting(), which read past $modSettings and its cache.

Two things the rollback does not cover, both documented in the class: DDL, since
MySQL commits implicitly on CREATE/ALTER/DROP; and anything happening in
another process, such as a request made over HTTP.

The three tests

• HarnessTest checks the harness itself — that the rollback really happens
  (one test writes, the next asserts it is gone), that $modSettings is
  restored, and that assertNoErrorsLogged() is capable of failing. A suite
  whose isolation quietly broke would not fail; it would start passing things it
  should not.
• ModSettingsTest pins the counter regression fixed in [3.0] Makes the settings counters increment on PostgreSQL too #9340.
  updateModSettings($x, true) emitted SET value = value + 1 against
  settings.value, a text column.
• SchemaTest compares Sources/Db/Schema/v3_0/ against the live database
  in both directions — declared tables and columns that are missing, and
  columns present that nothing declares. This is the drift AGENTS.md warns about,
  where a query naming a removed column fails at runtime only, and inside a
  background task retries forever.

Why both engines, concretely

Not thoroughness for its own sake. I reverted #9340's fix and re-ran:

	MySQL	PostgreSQL
with the fix	9 pass	9 pass
fix reverted	9 pass	5 fail

MySQL coerces text to a number and hides the bug completely. A suite that only
ever saw one engine would have proved nothing here. Worth knowing too: on
PostgreSQL a failed query poisons the rest of the transaction, so one swallowed
error turns every later query in that test into false — queryRow() reports
that as a readable assertion failure rather than a TypeError about argument #1.

Notes

• tests/bootstrap.php gains one line registering SMF\Tests\ for autoloading.
  The unit tests are each self-contained so nothing needed it before; anything
  sharing a base class does. Kept beside the two setPsr4() calls already there
  rather than adding an autoload-dev section only the suite would use.
• composer.json gains test-unit and test-integration alongside test.
• AGENTS.md is updated: it previously told contributors that Db::$db was
  simply out of scope.

Verified

• .docker/test.sh — 128 tests, 196 assertions, green on both engines.
• With Settings.php removed — 108 pass, 20 skip, exit 0.
• Regression proven in both directions, as above.
• check-signed-off.php, check-smf-index.php, check-smf-license.php,
  check-smf-languages.php and phplint all pass; shellcheck clean on all
  five .docker scripts.

Two commits that are not about testing

• Lets the section comment fixer place its own banners — I wrote the section
  banners by hand with the wrong asterisk count, so SMF/section_comments did not
  recognise them and added its own alongside. AGENTS.md already says not to
  hand-write these; this takes what the fixer produces.
• Removes the trailing tabs from a blank line in PM search — one character in
  Sources/PersonalMessage/Search.php, which I have otherwise not touched.
  c344b5c23 left a line holding nothing but three tabs. It has not turned CI red
  so far because the style workflow normally only checks the files a PR changes;
  it checks everything when composer.lock is in the diff, which this PR's
  does. Any other branch touching a dependency will hit the same thing, so it
  seemed better to fix than to work around. Happy to split it out if preferred.

Merge order

Merge #9326 and #9344 before this one. Both are contained in this branch, so the
diff shown here is theirs as well as its own; once they land and this is rebased on
release-3.0, what is left is tests/Integration/ and its bootstrap.

Issues References (Fixes|Related|Closes)

1. Depends on [3.0][Testing] Install the forum from the command line #9344 — needs .docker/install-forum.sh to have a forum to test.
2. Depends on [3.0][Testing] Add a PHPUnit suite for the parts that need no database #9326 — builds on the PHPUnit suite and its bootstrap.
3. Related to [3.0][Testing] Add a Docker development environment for MySQL and PostgreSQL #9317, [3.0] Makes the settings counters increment on PostgreSQL too #9340.