Checking that the platform supports foreign key is not the right case
here, because we should check if we can create them by themselves, after
the table is created. It is not the case for Sqlite.

Closes #3990

Bump cs libs

Upstream libs and tools have improvements that make them unnecessary.

[GH-1204] Add full support for foreign key constraints for SQLite

Use composer install in the Psalm job

Our policy states that we do not want a new release to suddenly affect
our builds.

Remove unneeded and risky stubs parsing

PHPStan already reads these stubs, but in a controlled and safe why. It
does not seem to read global constants yet, which is why we define
OCI_NO_AUTO_COMMIT if undefined. Might be unneeded in the future too.

This constant is not recognized by PHPStan, and besides the docs say
this:

> Prior to PHP 5.3.2 (PECL OCI8 1.4) use OCI_DEFAULT which is equivalent
to OCI_NO_AUTO_COMMIT.

See https://www.php.net/manual/en/function.oci-execute.php

Psalm 7

The mismatch is deliberate.

This part of the code is too convoluted for Psalm (and maybe for humans
too).

That method is already defined in the parent interface.

This was shown as an error by Psalm but should not be considered one
IMO. See https://github.com/vimeo/psalm/pull/3203

Stop relying on the master version of Psalm

Mark every exception as immutable

I just got bitten by a bug on master at
https://github.com/doctrine/sql-formatter/pull/43#issuecomment-615576443,
and there are good chances we have the bug on the Dbal too. Anyway, it
only makes sense to use the same version as in the vendors.

Reworked LoggingTest to be able to test Statement::executeUpdate()

Remove baseline

null contradicts the phpdoc