v0.15.5

Added

• Response calibration functions: Add configurable NumExpr response transformations as a dedicated calibration stage, separate from detector gain calibration.

Changed

• Parsing and data-object performance: Reduce allocations in LArCV cluster parsing and cache data-class reflection metadata used for field validation and serialization.
• Data-object type enforcement: Preserve declared array dtypes on assignment and reject undeclared attributes, keeping reconstructed objects internally type-consistent.

Fixed

• Field-corrected reconstructed positions: Restore temporary module translations before returning calibrated coordinates and propagate field corrections to every declared reconstructed positional attribute, including start, end, PPN, and vertex points.
• Mixed-precision reconstruction: Keep coordinate, endpoint, and direction dtypes consistent so Numba tracking and vertexing kernels do not fail on mixed float32/float64 inputs.

Full Changelog: v0.15.4...v0.15.5