Fix __call in PDO subclass breaking method calls

[iluuu1994]

Fixes GH-7803

Are changes like these acceptable after the feature freeze? The issue is marked as a feature, but that's mostly so that we don't back-port the fix and change longstanding behavior.

[Girgias]

Win test failure seems to be related but is quite odd:

Environment: THREAD_SAFE=0, OPCACHE=0, PARALLEL=-j2

========DIFF========
001+ Fatal error: Uncaught PDOException: could not find driver in C:\projects\php-src\ext\pdo_sqlite\tests\gh7803.php:11
002+ Stack trace:
003+ #0 C:\projects\php-src\ext\pdo_sqlite\tests\gh7803.php(11): PDO->__construct('sqlite::memory:', '', Object(SensitiveParameterValue), Array)
004+ #1 {main}
005+   thrown in C:\projects\php-src\ext\pdo_sqlite\tests\gh7803.php on line 11
001- __call: nonexistent
002- array(2) {
003-   ["CONCAT('123', '456', '789')"]=>
004-   string(9) "123456789"
005-   [0]=>
006-   string(9) "123456789"
007- }
========DONE========
FAIL GH-7803: PDO __call breaks method calls [C:\projects\php-src\ext\pdo_sqlite\tests\gh7803.phpt]

Otherwise I think it is fine to merge this after feature freeze as this is self contained and more a bug fix than anything else IMHO

[cmb69]

We need to explicitly declare the required extensions (pdo_sqlite is not loaded on AppVeyor by default):

Suggested change

	--FILE--
	--EXTENSIONS--
	pdo_sqlite
	--FILE--

[cmb69]

I'm not quite sure if we should fix this in this way. I'd very much prefer a more general solution like https://wiki.php.net/rfc/pdo_driver_specific_subclasses, but of course, it is not clear that we will have something like this in the foreseeable future. :(

[iluuu1994]

@cmb69 I didn't know about this RFC, I agree that solution would be preferable. I'll close this for now and reopen in case the RFC would fail.