Improve serialization robustness: schema preservation, triangulation, and checksum alignment

[afshawnlotfi]

Enhanced schema utilities and add earcut triangulation

• Schema Enhancements: Improved discriminator matching in SchemaUtils to validate all Literal fields present in both model and data
• Computed Fields: Updated serialization to include computed fields (bbox, render_order) when extracting BaseModels
• Triangulation Support: Added earcut library dependency for triangulating non-triangle faces in mesh processing
• TypeScript Updates: Enhanced mesh.ts and packable-worker.ts with triangulation capability

Packable schema handling and checksum improvements

• Schema Preservation: Added _original_json_schema to Packable class to preserve schema during reconstruction (for dynamic models with numpy fields)
• Fallback Mechanism: Added error handling in extract() to use preserved schema when model_json_schema() fails
• Checksum Refactor: Rewrote compute_dict_checksum() to compute SHA256 from compact JSON + asset bytes (removing orjson dependency)
• Type Safety: Improved schema resolution for untyped references (object/Any types)

Refined JSON schema handling and TypeScript alignment

• Better Abstraction: Extracted _get_json_schema() method for cleaner fallback logic with improved documentation
• TypeScript Parity: Updated checksum computation to match TypeScript implementation (compact JSON with (',', ':') separators and conditional asset bytes)
• Code Clarity: Enhanced normal computation in meshToRawGeometry() with clearer inline comments and improved formatting
• Edge Case Handling: Refined logic for detecting array references vs raw dict values with $ref keys

Overall Focus: These commits improve serialization robustness, cross-language consistency (Python/TypeScript), and support for dynamic/complex models.