Switch to using notebook %pip magic when in IPython env for check_and_install_missing_packages

[ianhelle]

In Azure ML, using pip in a subprocess or using "!pip" from the notebook causes cross-kernel problems for the conda environment.
Specifically if you !pip install a package in py36 conda environment, then switch to py38 environment you cannot either import the package, nor can you !pip install it. Pip finds the package in the py36 environment and claims it is already installed.
The package is however, not available to the py38 kernel.

Change the code in check_and_install_missing_packages to:

• detect if running in IPython
• use get_ipython().run_line_magic
• run %pip install pkg

[ianhelle]

Apparently I already implemented this!