Diffusion / flow specs (umbrella)

This is the curated public entrypoint for TorchLean's diffusion / flow spec layer.

It re-exports:

• a discrete VP schedule + forward noising (qSample),
• reverse samplers (DDPM and deterministic DDIM), and
• a continuous-time VP schedule + probability-flow ODE drift (pfOdeRhs).

All specs are scalar-polymorphic (Context α) so the same definitions can be reused for:

• runtime execution (Float, IEEE32Exec, NeuralFloat, …),
• proofs (ℝ), and
• verification backends (interval scalars, etc.).