Added
- Introduced 2D and 3D vector classes supporting NumPy and Awkward arrays.
- Added support for jagged Awkward arrays in vector operations.
- Implemented vector rotation methods in 2D and 3D.
- Introduced `backend_tan` for tangent computation.
- Added scalar input support to `backend_log` and `backend_exp`.
- Introduced `compute_mass` safeguard for negative mass squared.
- Added jagged array inner dimension consistency check in utilities.
- Added fine-grained caching system with performance instrumentation.
- Added LHCb-specific vector performance benchmarks.
- Added comprehensive benchmarking against the `vector` package and for LVec performance.
- Added `requirements.txt` file (generated via `uv`).
- Added `pytest` to development dependencies for `uv run pytest`.
Changed
- Updated Python requirement to 3.10+ due to `numpy` compatibility.
- Updated example documentation with real LHCb Open Data analysis.
- Extended README with 2D/3D support and installation instructions.
- Updated LHCb example to use internal `to_np` function instead of `ak.to_numpy`.
- Updated README to reflect LHCb benchmarks.
- Updated `pyproject.toml` with missing dependencies for `awkward`.
Fixed
- Corrected vector class names from `LVec2D`/`LVec3D` to `Vector2D`/`Vector3D`.
- Fixed multiple utilities and backend functions to better support scalar and jagged inputs.
- Improved robustness of `compute_mass` against invalid inputs.
Refactored
- Unified use of `backend_where` instead of `np.where` for consistency.
- Improved clarity in pseudorapidity calculation.
- Removed redundant `np.atleast_1d` calls in `to_ak`.
- Restructured caching logic for better modularity.
Removed
- Removed `uproot` from dependencies, as it is no longer required.