Add AI-assisted semantic normalization with provider failover and manual fallbacks

[Prateek32177]

Motivation

• Normalization across platforms is lossy and brittle because field names vary, so we need a layered approach that preserves raw payloads and lets developers declare the schema they care about.
• Provide an AI-powered extraction layer to map developer-declared semantic fields to platform payloads while keeping a manual fallback for reliability and migration scenarios.

Description

• Added semantic types and options in src/types.ts (SemanticNormalizeOptions, SemanticAIOptions, _semantic result shape and per-field meta).
• Implemented the AI extraction + fallback engine in src/normalization/semantic.ts with provider failover order (groq, cohere, openai, anthropic, google), dynamic module loading, JSON-parsing of AI responses, and confidence threshold handling.
• Made core normalization async by changing normalizePayload to async in src/normalization/simple.ts, and attached semantic results to normalized payloads under _semantic when normalize.semantic is provided.
• Updated verification flow in src/index.ts to await normalizePayload(...) so semantic extraction runs inside the verification pipeline.
• Added a regression test in src/test.ts to verify semantic extraction falls back to manual dot-paths when AI is unavailable/low-confidence, and added documentation and usage examples in README.md (API, provider env vars, example config).

Testing

• Built the project with npm run build (TypeScript compile) which completed successfully.
• Ran the test suite with npm test; all tests passed including the new semantic fallback regression test (Semantic fallback: PASSED).

---

Codex Task