Fix database settings

[dan-fernandes]

Fixes #1223

[danielballan]

Thanks so much @dan-fernandes! This is great, and very appreciated.

We observed a related recent regression in #1228. I suspect this will fix that as well.

I pushed some additional commits to your PR branch here fixing these failures:

FAILED tiled/_tests/test_sync.py::test_copy_internal - tiled.client.utils.ClientError: 409: /a http://local-tiled-app/api/v1/metadata/
FAILED tiled/_tests/test_sync.py::test_copy_skip_conflict - tiled.client.utils.ClientError: 409: /a http://local-tiled-app/api/v1/metadata/
FAILED tiled/_tests/test_sync.py::test_copy_warn_conflict - tiled.client.utils.ClientError: 409: /a http://local-tiled-app/api/v1/metadata/
FAILED tiled/_tests/test_sync.py::test_copy_error_conflict - tiled.client.utils.ClientError: 409: /a http://local-tiled-app/api/v1/metadata/
FAILED tiled/_tests/test_sync.py::test_copy_external - tiled.client.utils.ClientError: 409: /group http://local-tiled-app/api/v1/register/
FAILED tiled/_tests/test_sync.py::test_copy_search_results - tiled.client.utils.ClientError: 409: /a http://local-tiled-app/api/v1/metadata/
FAILED tiled/_tests/test_sync.py::test_copy_items - tiled.client.utils.ClientError: 409: /a http://local-tiled-app/api/v1/metadata/
FAILED tiled/_tests/test_sync.py::test_copy_dict - tiled.client.utils.ClientError: 409: /a http://local-tiled-app/api/v1/metadata/

I think your last commit, introducing use_cached_database, attempted to fix this issue but in doing so created a new issue, as a distinct in-memory database would be created on every request. My approach reverts that commit and instead uses two distinct (file-based) databases in the test. (If we prefer, we could specify distinct in-memory databases using a URI like sqlite:///file:SOME_UNIQUE_NAME?mode=memory&cache=shared&uri=true.)

Edit: In fact, I did add a commit to use this "named memory" approach. It's a lighter-touch change than introducing file-based databases, and it may run a little faster on CI, which has very slow disk.

[danielballan]

Minor items

[danielballan]

I introduced new utility function, is_memory_sqlite, with tests. This is an improvement—more robust and consistent—but I am still seeing this failing test:

FAILED tiled/_tests/test_validation.py::test_unknown_spec_permissive - sqlalchemy.exc.OperationalError: (sqlite3.OperationalError) no such table: nodes

It appears to be flaky, involving a race condition because locally it will sometimes pass.

[danielballan]

I think those last commits did the trick. My train is almost home, so I'll hand off to 🇬🇧 here.

[dan-fernandes]

As mentioned in #1223, I've gone back on caching all in-memory databases, and instead given the authn database a unique in-memory, cached URI. This more closely resembles the expected behaviour of SQLAlchemy and avoids weird workarounds.

Also, minor changes.