Changed methodology for benchmarks

* Usage of pytest-benchmarks is dropped since it call tests
  sequentially which very long time to complete, instead original
  tavern tests are used
* Time measuring is done via durations switch to pytest and tavern
* During benchmarking validate_response.py is disabled using nevly
  introduced env variable
* Each run creates xml junit file with time data
* After all runs data from files are combined and report file is
  generated
* Tests above threshold are marked with red