Description:
Create a library of pre-built, reusable requirements for common guardrail patterns that integrations can use out of the box.

Key deliverables:

• mellea/stdlib/requirements/guardrails.py
• Common patterns: no PII, no harmful content, factual grounding, format compliance
• Easy composition via RequirementSet or similar
• Clear documentation and examples

Acceptance criteria:

• At least 10 pre-built common guardrails
• Each guardrail has clear documentation
• Guardrails are easily composable
• Performance is acceptable (lazy evaluation where possible)

Pre-built guardrails to include:

• no_pii - Rejects outputs containing PII patterns
• no_harmful_content - Rejects harmful/toxic content
• json_valid - Ensures output is valid JSON
• max_length(n) - Enforces maximum character/token length
• min_length(n) - Enforces minimum character/token length
• contains_keywords(keywords) - Requires specific keywords
• excludes_keywords(keywords) - Forbids specific keywords
• matches_schema(schema) - Validates against JSON schema
• is_code(language) - Validates output is valid code in language
• factual_grounding(context) - Checks output is grounded in context