StreamingOnnxReader

Streaming ONNX reader that parses metadata without loading tensor data.

Memory usage is proportional to metadata size (~1-10 MB), not file size (100+ GB). Individual tensors can be loaded on-demand via loadTensorData.

This enables parsing of very large ONNX model files without requiring the entire file to fit in memory.

Usage:

StreamingOnnxReader.open(source).use { reader ->
    // Access metadata immediately - only metadata loaded
    println("Tensors: ${reader.tensors.size}")
    println("IR Version: ${reader.irVersion}")

    // Load specific tensor when needed
    val weights = reader.loadTensorData("conv1.weight")
}