Summary

Implements the complete data-layer file loading pipeline for meshscope (Feature 1 of 10 in the MVP Cutline). Loads STL (binary/ASCII), OBJ, 3MF, and PLY mesh files into an in-memory MeshDocument with full validation, error handling, and computed metadata. No UI — viewport integration comes with Feature 2.

What's included

• Exception hierarchy (MeshLoadError → FileValidationError / MeshParseError subclasses) with user-facing messages ready for UI display
• Immutable data model (BoundingBox, MeshMetadata, MeshData frozen dataclasses) with typed numpy arrays (mypy strict clean)
• UndoStack ring buffer shell (populated by Features 7-8)
• MeshDocument mutable session wrapper with deep-copied original mesh for reset
• Path validation (extension, existence, readability, 500MB size limit)
• load_mesh() pipeline: validate → detect format → pre-parse STL validation → parse via trimesh (explicit file_type, never generic load()) → Scene handling for 3MF → convert to data model → OBJ unsupported directive warnings → unit mismatch detection → structured logging
• Test fixtures for all 4 formats (valid + invalid) with a regeneration script
• 61 tests covering success paths, warnings, error conditions, and data integrity

Design spec deviations (deferred)

• source_format stores "stl" / "ply" rather than sub-types ("stl_binary" / "ply_ascii"). Plan tests assert the simpler form. Can add sub-type detection when UI needs it.
• parse_mesh is inlined in load_mesh rather than extracted as a separate function. Can extract if reuse is needed.

Test plan

• 61 unit tests pass (pytest tests/ -v)
• mypy strict clean (mypy src/meshscope/)
• ruff clean
• All pre-commit hooks pass (semgrep, gitleaks, ruff, trailing whitespace)
• Verify CI passes on this PR

🤖 Generated with Claude Code