Add calibrated redox screening analysis

[galjos]

Summary

• audit canonical structures and duplicate conflicts
• fit grouped DFTB-to-DFT corrections with uncertainty and extrapolation flags
• select non-dominated candidates with Pareto objectives
• document the workflow and anthraquinone benchmark

Validation

• 453 tests passed, 16 optional-backend tests skipped
• 98% total coverage; new module 100%
• strict documentation build and package lint passed
• grouped benchmark reduced MAE to 29.2 mV for E1 and 35.4 mV for E2

[galjos]

Closing because learned correction, descriptor modeling, and candidate ranking are outside the package scope.