fix: schema lock for batched per-atom columns

[Ramlaoui]

Summary

• derive schema slot bytes separately from batch payload stride for per-atom batched columns
• keep record slicing behavior unchanged
• add regression coverage for variable-size custom atom_properties and float64 per-atom builtins
• bump package/release metadata to 0.3.1

Closes #33

Tests

• cargo fmt --check
• INPUT_TAG=v0.3.1 python release version validation snippet from .github/workflows/release.yml
• cd atompack-py && uv run --no-sync maturin develop
• cd atompack-py && uv run --no-sync pytest tests/test_database.py -q -k "variable_size or custom_properties or float64_geometry"
• cd atompack-py && uv run --no-sync pytest tests/test_database.py -q
• cd atompack-py && uv run --no-sync pytest -q

Note: local commits are unsigned because the GPG pinentry prompt timed out in this non-interactive session.