Content safety check

Score a text for unsafe content across seven categories — toxicity, profanity, harassment, hate_speech, self_harm, sexual, violence — each as a 0-1 probability. Returns scores (all seven), flagged (whether any category reached the threshold, default 0.5) and flagged_categories. With spans=true, also returns spans: the specific offending passages, each with its text, start/end offsets into the submitted text, per-category scores and flagged_categories — use it to highlight, quote, or redact just the violating passage instead of rejecting the whole text.