Generate a Docker container within the CI build

[algattik]

Resolves #141

Changes:

• Generate a Docker container within the CI build, tagged with a checksum of the Dockerfile and requirements.txt files so it can be reused across run. Store this Docker container in the AML ACR. Use this Docker container for running unit tests.
  

• Other jobs still use the mcr.microsoft.com/mlops/python:latest image as before. This allows derived projects to install "large" ML frameworks in the CI build, without making subsequent jobs needlessly slow to start.

• Allow developers to lint & test locally in a single command: streamlined use of flake8 and pytest within tox.

$ tox
python run-test-pre: PYTHONHASHSEED='1159078597'
python run-test: commands[0] | flake8 --output-file=lint-testresults.xml --format junit-xml
python run-test: commands[1] | pytest tests/unit --cov=code --cov-report=xml --junitxml=unit-testresults.xml
=========================================================================== test session starts ===========================================================================
...
tests/unit/code_test.py .                                                                                                                                           [ 20%]
tests/unit/data_test.py .
...                                                                                                                                        
---------- coverage: platform darwin, python 3.7.6-final-0 -----------
Coverage HTML written to dir htmlcov
Coverage XML written to file /Users/algattik/GitHub/MLOpsPython2/coverage.xml

====================================================================== 5 passed, 4 warnings in 6.34s ======================================================================
_________________________________________________________________________________ summary _________________________________________________________________________________
  python: commands succeeded
  congratulations :)

• Added code coverage report. Required adapted pytest commandline (now in tox.ini) and installing .NET Core as for security reasons, the AzDO PublishCodeCoverageResults does not publish user-generated HTML reports anymore.
  

• Linting fixes in some .py scripts to avoid CI build failing.

• Some simplifications in the build pipeline YAML.

Notes to MLOpsPython repo maintainers:

• The pipeline building the mcr.microsoft.com/mlops/python:latest image was moved from environment_setup/toenvironment_setup/mlops-image`. The intention was to make it clear to users that there are two separate Docker images. The production AzDO pipeline that maintains the mcr.microsoft.com/mlops/python:latest image will have to be updated.
• The mcr.microsoft.com/mlops/python:latest image was updated to include only the modules needed for running the ml_service scripts, i.e. removing numpy, pandas and scikit-learn. This may break existing users and should be evaluated, we can also put those modules back. Perhaps the best solution is to abandon mcr.microsoft.com/mlops/python:latest and just build that image in the project's ACR, like the CI build image?
• Suggesting to adopt a fixed tagging scheme for the mcr.microsoft.com/mlops/python repo for reproducibility. Also ideally fixing the AML SDK version.

[eedorenko]

Requested changes

[algattik]

Superseded by #150