Repeated tests on same machine might pass when they should fail

I haven't caused the break myself, but looking at [this test](https://github.com/canonical/serialized-data-interface/blob/09c66d603bcb6d6ca7ef68f5676d1a0b3d53d948/test/unit/test_utils.py#L14) I think because of the caching it is possible to break the fetching of remote schemas without breaking the test by:
* run the test when everything works on a schema url. schema file is cached locally in `/tml/hash`
* break the code that fetches remote schemas
* run the test again on the same url. This will hit the cached file and pass even though it can't fetch it

Some easy fixes I think would either be testing on the remote-fetching more directly, or adding a `ignore_cache` arg?