OnnxModelParser

ONNX model parser extending BaseModelParser. Provides metadata-first parsing of ONNX model files using SKaiNET's ONNX I/O capabilities.

This implementation supports two modes:

• Streaming mode (JVM): Uses RandomAccessSource for memory-efficient parsing. Only metadata is loaded (~1-10 MB), tensors loaded on-demand via loadTensorData.

• Legacy mode (JS/Native): Falls back to pbandk which loads full file.

Usage:

val parser = OnnxModelParser()
parser.parseMetadata("model.onnx")

// Check if streaming mode is available
if (parser.isStreamingMode) {
    // Load specific tensor on demand
    val data = parser.loadTensorData("conv1.weight")
}

// Always close when done
parser.close()