Since our current implementation is pretty much a stub anyway, this is fine.
That said, the tests are not exactly interesting (we only try setting it to MEDIA_TIME, which is already the format, and NONE, which is kind of special). As usual, I ended up writing more tests to validate the implementation, and found, interestingly enough, that while the filter graph is fine with changing the time format while running, it does not expect the same of its filters, and will actually stop and restart itself if you try to actually change the time format to something interesting. Rudimentary tests attached here.