Module model 

Inference node (1×in → 1×out) that runs a generic ComputeBackend model.

Design

• No dynamic dispatch: backend and model are monomorphized by generics.
• No unsafe in the hot path.
• Batching:
  • no_std / no-alloc: stack-bounded batching up to MAX_BATCH.
  • alloc: uses Vec for flexible batch sizing.
• Queues/telemetry are accessed only via StepContext.
• Zero-copy preferences are expressed through PlacementAcceptance.

This node delegates inference to the model (infer_one / infer_batch), and pushes outputs directly to the provided output edge. It never copies unless required by payload semantics or batch buffering.

Structs

InferenceModel

Generic 1×1 inference node for any backend (dyn-free).