Use new OpenVINO Python API features

[jiwaszki]

As OpenVINO 2023.1 came out, there are few new features in Python API which project can utilize.

• There is no need to use infer_new_request. Class CompiledModel handles creation of new request and reuse it between runs, saving some time.
• Better handling of data, which includes direct use of Pytorch Tensors.
• Additional performance benefits that come with memory sharing on inputs and outputs, thus reducing copies to minimum (share_inputs and share_outputs flags).
• openvino.runtime can be easily replaced with openvino flat namespace.

I spend some time benchmarking the solution, I will be happy to share results on request. I can also assist in case of any questions about other OpenVINO features:)

@zhuzilin thanks for creating this solution, always happy to see OV in action!