Unified entry point for AI-accelerated CFD predictions.  Translates high-level mode requests
into surrogate + corrector pipelines and always returns a canonical structure:
  {:flow-field … :constraints … :confidence … :metadata {:mode … …}}

Shared data structures, validation helpers, and fixtures for computational fluid dynamics
requests.  Every solver (classical, reduced-order, or ML surrogate) consumes the same
geometry/environment/platform maps so higher layers can reason about environmental flows
without caring about the underlying numerical method.

Helmholtz–Hodge projection for surrogate flow fields.  Uses a conjugate-gradient Poisson solve
to enforce near-zero divergence, applies boundary conditions, and limits energy so downstream
planners receive coherent, physically plausible velocities.

Debris/sand flow helper wrapping the surrogate + corrector pipeline for near-ground transport.

Maritime flow helper: prepares shallow-water geometries, water environments, and runs the
surrogate + corrector pipeline.

High-level plume modeling helper that wires geometry/environment normalization into the surrogate
and corrector pipeline.

Spatial surrogates for fluid prediction.  Provides deterministic analytical fallbacks and a hook
for plugging in future ML models.  Returns structured flow fields plus provenance metadata so
downstream correctors and translators can reason about confidence.