Add a composable profanity language profile API

[alyldas]

Problem

Language-specific profanity policy should compose into one read-only profanity filter without registering each language as a separate pipeline filter or recompiling maintained dictionaries.

Acceptance Criteria

• Add explicit, symmetric language entrypoints with declarative profile identity.
• Compose multiple profiles into one filter with match provenance and per-profile policy.
• Preserve diagnostic match evidence while merging masking and scanner ranges.
• Make dictionary normalization explicit and keep UTF-16 source-range invariants.
• Preserve legacy mutable factories and add an object-based custom-term factory.
• Support allocation-aware composed scanner streaming.
• Document and validate the public API, distribution entrypoints, and benchmark coverage.

No publish, no merge, no tag/release.