Run tests in subprocess instead of reloading robot code

[virtuald]

Managing vendor code is becoming untenable, maybe we should launch the robot code from scratch each time instead of reusing the same interpreter.

This wasn't previously possible, but now that things are launched using python -m robotpy it is.

• https://www.chiefdelphi.com/t/pathplannerlib-test-failures/452940
• #229
• Other things