Add remote session functionality for tvm_launcher.py

[dsbarinov1]

Motivation: until the current patch, tvm launcher was not able to work with RPC session, hence validating models on remote devices was impossible. In addition, it was also impossible to validate models on the GPU.

This patch is intended to bring this functionality with two additional flags "device" and "session", which are responsible for selecting "cpu" or "gpu" devices and "local" or "remote" sessions respectively.
At the moment, remote session and GPU device selection are implemented for android devices.

[dsbarinov1]

Motivation: some neural network models require compilation with the "Virtual Machine" module to work properly with them. An example of such a model is "ssd-resnet34-1200-onnx".

This patch is intended to integrate support for the "Virtual Machine" module when running "Accuracy Checker" with the "tvm launcher" by adding an additional "vm" flag that allows you to specify with "true" or "false" whether "Accuracy Checker" inference should be started with "Virtual Machine".
In addition, at this point it is necessary to specify the "outputs" parameter, which will contain the names of the output layers of the model.

Important: at the moment, all parameters need to be written explicitly in the configuration file "accuracy-check.yml" in fields that are at the same level as the path to the model in .so format or, for example, at the sane level with "inputs" parameter.

[dsbarinov1]

@KJlaccHoeUM9l could you please review my updates? :)

[dsbarinov1]

Motivation:
[Hexagon][Accuracy Checker] Integrate Hexagon target into Accuracy Checker: https://octoml.atlassian.net/browse/DEELVIN-372