I strongly suspect that we're currently running way too many
concurrent tests on some CI machines, which end up causing
timeouts.

Since !396 we now have functional tests that use more than a single
entry point.
We need to account for those, otherwise we might consider the discovery
to be completed too early and the tests would fail since we didn't
discover all the media that we were expected to

The tests will remove their folder upon successful runs, so any
remaining test.db file should be a failed test.
This will hopefully help us to investigate the sporadic test failures

We want to ensure the test folder gets removed when the test completes
properly, which, on windows/wine at least, requires all handles to be
closed.
Additionaly, the code to handle the test folder deletion is already
present in the parent class

Which now contains some files to test for the regression described
in #322

Fix #322

This was a perfect duplicate of Folder.DiscoverBanned, and wasn't
properly testing for banning an already discovered folder

Regression introduced in ad6fac64
Fix #319

All ban/unban methods are taking an MRL and not a path

So that they don't use the same database path.
Additionally, this can help us to package the failing tests database for
inspection after a failure on the CI

This is hardly usable in some test contexts

So that we now have a common initialization path for all tests

We don't access those directly but through a MediaLibrary member
function, so they can be private

This simplifies the initializations, and prevents having 2 callback
instances in tests that need a specific callback type with extra
functions

Which doesn't test something in particular, but which tries to reproduce
a more real-life like setup, as opposed to most test cases which only
have a single folder being discovered.
This would have caught the regression caused by !392 and fixed in !395

This also drops the --testdir parameter which was never used since it
was introduced

Since the function isn't recursive anymore, we are probably better off
skipping to the next folder in the discovery stack rather than aborting
the discovery entirely.
This is a good indication that our tests are a bit too simplistic

subFolder is a const& and can't be moved, so let's not pretend that's
what we're doing

So that the logs don't show banned folders as being discovered