compute_question_set_hash raises UnicodeEncodeError for a question with a lone-surrogate escape #757

compute_question_set_hash in dokimi/dokimi/orchestrator/pipeline.py:35 joins the sorted id:question fingerprints and encodes them with .encode("utf-8"). A questions file with a \ud800 or \udcff JSON escape loads as a lone surrogate. Repro: compute_question_set_hash([{'id': 'q', 'question': json.loads('"n\\ud800"')}]) raises UnicodeEncodeError: 'utf-8' codec can't encode character '\ud800' in position N: surrogates not allowed. Fix: .encode('utf-8', 'surrogatepass'), as #720 does for the other digests of parsed data; the hash of every question set that encodes today is unchanged.

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