Schema sqlalchemy engine caching is per-instance

[eito-fis]

Describe the bug
The _sa_engine property of the Schema model is cached using the @cached_property decorator. @cached_property caches on a per-instance basis, which means the engine is not cached across different instances of the same schema object, leading to the engine being created many times.

For example, each table must access it's schema's engine which is then cached on that table's instance of the schema. However, as the engine is only cached on that instance, another table trying to access the same schema's engine will also create a new engine. As a result, something like the table list endpoint will create an engine for every table in the database.

Expected behavior
We should cache the schema's engine object at a higher level so it persists across instances.

Additional context
Issues with too many connections, specifically the too many clients error, is blocking backend work that adds additional tests (see #265, #268). A naive caching implementation:

_engine = [None]
class Schema(DatabaseObject):
    database = models.CharField(max_length=128)

    @property
    def _sa_engine(self):
        if _engine[0] is None:
            _engine[0] = create_mathesar_engine(self.database)
        return _engine[0]

does resolve the error on #265. Unfortunately, I haven't been able to consistently reproduce the error as a single test.

I have also talked to @mathemancer about opening a broader issue on the where engines should be created. If this overlaps with that topic, I'm happy to close.

[kgodey]

Added this to milestone 2 because it seems important to resolve ASAP.

[eito-fis]

Note: We have a temporary fix in place as of #265 which should be removed when this issue is handled.

[dmos62]

As @eito-fis mentioned, this has been solved, and the solution has proven somewhat stable. There was a bug, where a cached engine might unexpectedly break when dropping and recreating a database, but that's fixed in #1222.