* Move tests to the right test file (e.g. tests for
  `dotenv.main.get_key` go to `tests.test_main.py`, not
  `tests.test_cli.py`.
* Check logged warnings for more functions.  For instance, that
  uncovered unexpected extra warnings when passing a non-existent file
  as argument.
* Remove tests for properties already checked at a lower level.  For
  instance, it's unnecessary to call `get_key` with weird file content
  because that's already tested with the parser.
* Follow the Act/Arrange/Assert (AAA) pattern.  This clarifies what is
  tested in each test function.  For instance, using this, we test only
  one thing at a time, which makes the test easier to read and potential
  failures easier to interpret.

* Fix PyPy jobs in CI

* Add test cases for `dotenv.set_key`

* Escape values before quoting in `dotenv.set_key`

This also simplifies tests and adds test cases for `set_key`.