fix(output_parsers): parse trailing category tokens for safe verdicts in is_content_safe

[krishna11-dot]

Summary

Fixes #2174

Problem

is_content_safe() in nemoguardrails/llm/output_parsers.py only extracted
trailing policy-category tokens (e.g., S1, S8) when the response prefix
was "unsafe". The "safe" branch unconditionally returned [True], silently
dropping any trailing category information.

Fix

• Added _parse_safe_violations() helper that mirrors the existing
  _parse_unsafe_violations() logic
• Updated the "safe" branch in is_content_safe() to call it
• Re-enabled the previously commented-out assertion in
  test_content_safety_input_with_is_content_safe_parser_safe_with_violations

Testing

All 36 tests in tests/test_content_safety_integration.py pass.