ai-sherpa/gp-fvm-vs-pinn-pde-surrogates-repro

claim 3: rebuild at the paper's full 8,857-dim mixed-functional state; PINN at the paper's own 4x128/2x64 + 30k-iter budget; decompose the previous 0.6012 to an inflow-boundary bug

rework(claim 6): GP-FVM extended to the coupled shallow-water system - propagation + shoaling verified

Update logbook: Repro - Scalable Bayesian Inference for Nonlinear Conservation Laws (GP-FVM)

initial commit