FIX recognize Azure SelfHarm in scorer evaluation

[Imran Ahamed (immu4989)]

Description

Azure Content Safety returns SelfHarm, which is not recognized by HarmCategory.parse_many. Scorer evaluation therefore rejects it against the registered self_harm dataset category, both when it is the only score and when selecting from all four Azure scores.

Add the explicit selfharm alias for SELF_HARM. This completes the remaining Azure case after #2575 while preserving strict rejection of unrelated categories.

Related to #2570; reproduction and proposed scope: #2570 (comment).

Tests and Documentation

• Parameterized regression coverage for all four registered Azure categories, with single-score and multi-score results. Both SelfHarm cases fail before the fix and pass afterward.
• Evaluator, Azure Content Safety scorer, and harm-category unit tests: 118 passed.
• All applicable pre-commit checks passed, including Ruff and type checking.
• No API calls or dataset/metrics changes; no documentation changes needed for this alias correction.