Resolves #106 by blanket ignoring dtype assertions when what's expected doesn't exist, e.g. the default uint. Couldn't introduce a warning easily as I forgot we're using Hypothesis, so opted to do this completely silently 😅

Also ignoring dtype assertions (as well as shape assertions) before testing values isn't ideal, as it can obfuscate the core problems. Fortunately for test_prod I can't see this being a problem with the PyTorch scenario, although other partial dtype scenarios could be problematic... but don't seem to exist!