Refine setup process 2

[Jiahang Xu]

• support batch mode in cmd line auch as nn-meter --onnx <file-or-folder> --predictor <predictor-name>
  • in code: predictor-name; in doc: hardware; in readme: device + inference framework
  • refine integration-test by batch mode
• complete IR model test
  • install Ubuntu to support nni development and install the latest nni package
    • reset python env in Ubuntu
    • nni install
    • test nni multi-trail.py in nni
  • fix nni-ir model in nn-meter
  • refine readme.md and give an instruction of test
  • report an issue in nni found when testing nn-Meter
  • add the integration test of nni-ir graph (found that the current nni does not support nn-Meter module, maybe we should waiting for the next release of nni)
• refine the API and model type list in nn-meter.py
  • change model type of --torch to --torchvision
  • change model type to --nni-ir and --nnmeter-ir
  • predict torch model by calling its name
  • support batch mode for torch model
  • add --torchvision test in integration_test_torch.py (intend to run parallel to save time)
  • refine readme.md
  • change torchvision type to torch considering the nn.Module in python binding
• add a new usage --getir to get nn-meter ir model for tensorflow and onnx model
  • --getir usage testing
  • add --getir usage test in integration_test.py
  • edit README.md
• edit the API in NNI (remove default config)
  • add nn-meter in related projects of nni
  • change 'nni' to 'nni-ir'
• arrange docs
• public
  • open a PR to refresh the data release link in config
• add hardware device attribution in config/predictors.yaml, and refine the hard code in predictor loading (category: cpu)
• add cache in integration test

[Yuqing]

• split the integration test into 4 .yml file
  this item could be ignored if we support batch mode in command line and the integrated test takes an acceptable time consumption
  here is a batch mode example
  nn-meter --onnx --predictor <predictor-name> <file-or-folder>