Implementing with PDO backend. Fixes #9

[rimas-kudelis]

This (hopefully) fixes #9, albeit with some caveats. Quoting the README, here are the missing features:

• Support for last insert id. The pdo-firebird extension doesn't support this natively. You implemented it by using a query, but it requires the name of a sequence as an argument. Not sure if implementing it in PDO this way would be much of a win over not implementing it at all. Futhermore, I'm not sure if it works within a transaction (I think it was failing for me with a "transaction is already active" or similar message, but that was before I actually fixed transactions, so it's probably not an issue anymore.
• Support for different transaction isolation levels. It appears that to implement this, we'd have to handle transactions on our side without relying on PDO. Your original Connection class was mostly about it, but this was a bit too much for me, so I decided to just go with the default isolation level for now.
• Transaction lock timeouts and waiting. Same issues as above.
• Autocommitting changes from last queries during object destruction, if they were not made within an explicit transaction. Again, the test just kept failing and both my patience and time were running out. But honestly, I'm just not sure why it wouldn't work out of the box when autocommit is on.
• Quoting queries without calling the constructor first. That's just not how PDO operates, and likely for a good reason.
• Generating and returning a DSN. This is a protected method of the driver now.
• Sanitization of options passed to the driver. I just rely on PDO to do the right thing.
• Support for Firebird SQL dialects. The underlying PDO driver claims to support dialects 1 and 3, but tests for dialect 1 just kept failing, even though the DSN I generated did include dialect version. I ended up just deleting the test. ext-pdo-firebird only supports dialects starting with PHP version 7.4, so this driver will also only support it starting on PHP 7.4 and up. Futhermore, only dialects 1 and 3 are supported according to PHP documentation, so even though the test for dialect 2 passes, I'm not sure it always will. And dialect 0? I guess there's no such thing. The test for dialect 0, which expected it to be treated as 3, failed, so I assume in PDO it falls back to 1 instead of 3. Either way, I guess it shouldn't be used or tested against.

I also removed some assertions about internal class architecture from the tests. Swapping the backend ruined those assertions. I think we're supposed to test how the unit works, as opposed to how it is constructed.

[rimas-kudelis]

BTW I also merged your 2.7 branch back into master.
Also, 52cf95a might be worth cherry-picking into 2.7 as well, since you broke symlinks you had committed previously. Although I don't think you should even commit them: they seem to be created automatically during composer install.

[rimas-kudelis]

The PR was closed because I deleted my fork, but I think you can still reopen it if you feel like it.