Add ir_version parameter to `create_onnx_model` function

Context

cf. discussion in aidge_core#308 (comment 4428218)

Would allow the benchmark script to specify an IR version when calling create_onnx_model() in main.