In `test_irc_utils` suite, rewrote `test_safe` and `test_safe_bytes` to
generate a bunch more permutations of invalid characters automatically
rather than writing them all out. The horse is definitely beaten *far*
past death at this point... but the tests sure are thorough(bred).

Also renamed `test_safe_null` -> `test_safe_none` to avoid confusion;
`test_safe` and `test_safe_bytes` are the cases that check behavior with
null characters/bytes. Python's docs technically do refer to `None` as
the "null object" once or twice, but most of the docs (and most of us
devs) simply call it `None`.

Using `pytest.mark.parametrize()` instead. Two test cases become 12.

Note: Yes, it is possible to have pytest treat each tuple generated by
`itertools.permutations()` as a single argument. Doing so yields useless
generated test-case names like `test_safe[seq0]`, though. Using three
separate test parameters of simple types (`str` or `bytes`) yields more
informative names like `test_safe[\n-\r-\x00]`.

Link to RFC explaining why CR, LF, and NUL are prohibited, and annotate
the version in which NUL stripping was added.