feat: add auto into --runtime or alike to auto select best suitable runtime

[xieofxie]

It is easy to tell from ort (an onnx file) and ort-genai (a folder with genai_config.json)

As a follow up of #1015

Folder itself is not enough if we support transformer style inference, it is still using ort not ort-genai

[xieofxie]

perf model input matrix

model flag input	backend	support input
onnx	AutoModel	random input
onnx	OpenVINO ?	random input
multiple onnx files (composite model)	AutoModel	random input
onnx folder with LLM config.json (*) (exported via ry)	AutoModel	random input ? / prompt
onnx folder with LLM config.json (*) (exported via Mxx)	?	random input ? / prompt
folder with genai_config.json	genai	prompt

*: or task + hf id ?