Backend predict hook

[dimdano]

Move the existing C-simulation logic into an internal _predict() method and update the public predict() API to optionally dispatch to a backend-provided predict() implementation. For example, this might be useful for external backends (e.g., AIE) to implement accelerator-specific execution while preserving the standard hls4ml prediction interface.

Type of change

• [x] New feature (non-breaking change which adds functionality)

Checklist

• yes all