Backend Contracts

Contract-carrying backend vocabulary for TorchLean runtimes.

The semantic graph and specs stay in Lean. Fast providers such as native CUDA, LibTorch, ATen, cuBLAS, cuDNN, or cuFFT enter through named capsules that record shape/layout/value/VJP contracts and an explicit trust level. The planner consumes these capsules under an ExecutionConfig and selects an admissible execution provider for each operation or fused operation group.