Atom id and the retention cache key raise UnicodeEncodeError for an atom statement or artifact text with a lone surrogate #758

Atom.__post_init__ (dokimi/dokimi/novelty/atoms.py:120) hashes self.statement.strip().encode("utf-8") when no id is given, and retention._key (dokimi/dokimi/novelty/retention.py:258,260) hashes the statement and the text the same way. A statement parsed from a model reply or an atoms file can hold a lone surrogate from a JSON escape. Repro: Atom(statement=json.loads('"n\\ud800"'), question='q', answer='a') and retention._key(Atom(statement=<that>, question='q', answer='a', id='a1'), 't') each raise UnicodeEncodeError: 'utf-8' codec can't encode character '\ud800' in position N: surrogates not allowed; the same happens with a surrogate in the text argument (line 260). Fix: .encode('utf-8', 'surrogatepass'); the id and the key of every statement that encodes today stay the same, which matters because cached verdicts are keyed on them.

Found while fixing #720. The fix's PR is #761.