Add auditable Jacobian regularization searches

[charlesmartin14]

Summary

• split the notebook-free analysis into four visible, resumable CLI stages
• add a five-point Tikhonov resolvent grid with an auditable min-valid-KS selector that never targets alpha=2
• add frozen-rank optimal-MP signal-space projector and log-Gram Jacobians, retaining the Feshbach collapse control
• expand materialized quotient representatives to Gram, Feshbach, and calibrated-MP parameter scans, all analyzed by WeightWatcher raw and clip_xmax
• add live progress/ETA JSON, separate optimizer plots, complete CSV inventories, and extensive shareable HTML documentation
• make report generation fail on missing local-Jacobian, Tikhonov, quotient, flow, or transport units

Validation

• Python compilation and shell syntax checks passed
• git diff --check passed
• exact ECS/Tikhonov/MP algebra smoke passed on synthetic wide matrices
• all expanded quotient-profile algebra checks passed
• existing 21 dependency-light weight-quotient tests passed
• full synthetic end-to-end report generated with 38 Jacobian trajectories, 200 Tikhonov candidates, 960 WeightWatcher quotient fits, 216 flow fits, and 36 transport rows
• visually inspected optimizer-specific Tikhonov output

Full PyTorch/WeightWatcher execution remains for the rg-muonclip-run environment on the saved short-100 checkpoints.