Performance Optimization for computing id hashes.

[beberlei]

In 09b4a75 we introduced an API UnitOfWork::getIdHashByIdentifier that added a significant amount of function calls over the previous inline implode to support backed enums as identifiers.

This PR moves ID hashing into a strategy pattern, moving the object to ClassMetadata instances of each entity, getting the previous performance almost back (+ the dynamic call).

I hope this will significantly improve performance of the 80% identifier that is a simple, single numeric identifier.

Here are the results from two traces compaed with Tideways using the lazy ghost performance project discussed in #11087 that runs EntityManager::getReference, hitting the ID hashing code significantly for 100.000 objects.

Old Call	Old Time	New Call	New Time
UnitOfWork::getIdHashByIdentifier	46.75 ms	SingleElementIdHashing::getIdHashByIdentifier	15.3 ms
UnitOfWork::tryGetById	33.01 ms	UnitOfWork::tryGetByIdWithClass	16.1 ms

While not life-changing, this hit accrues, especially in test-suites working with fixtures.

[beberlei]

Closing in favor of #11332