Standardize Python import ordering

[dbutenhof]

PBENCH-893

The ordering of Python import statements is a frequent topic of discussion during PR reviews. It's often not easy when writing code to determine whether a given module is "standard library" or not, and it's easy to accidentally add a new symbol out of order.

During a recent review, the isort utility was mentioned. I gave it a try but had some difficulty configuring it to be compatible with our black (despite the built-in "black" profile), and gave up.

As an excuse to put off dealing with the necessity of breaking my development VM (and Varshini's dashboard testing), I decided to experiment a bit more to see if we can solve this issue "once and for all" as we've done with source format style in general by adopting black.

I've now got a configuration (defined in our pyproject.toml file) that causes isort to generate output that makes black happy. (This was not trivial, sadly.) I've also replicated the black exclusion list, as that seemed to be appropriate.

I've added isort along with black and flake8 to the lint tests in our build.sh script.

NOTE: see https://pycqa.github.io/isort/docs/configuration/git_hook.html for instructions on adding isort to your git precommit hook. I've integrated the isort extension (and configuration) into my vscode. (Which will be much more useful once we get all modules to a standard baseline.)

isort changes 132 files in the tree, which are included in this PR. This includes obsolete legacy test files, which I decided weren't worth the effort of excluding.

[npalaska]

This is good, so we can just run isort on the python file after adding new imports. We just have to run isort before we run black.

[portante]

See #2988 for pre-commit hook work.