More idiomatic and consistent unit tests

[kba]

When we began OCR-D/core, we targeted only unittests with some enhancements over time in tests/base.py. To run the tests, we're actually using pytest, though.

We currently have ~300 tests which use various mechanisms to achieve the same task, often with nested context Managers (with TemporaryDirectory() as tempdir: and such).

@M3ssman suggested embracing pytest's features to a larger extent. For example, pytest offers decorators to achieve much of what we currently implement ourselves. For clarity and maintenance, we should refactor the test suite in a more consistent style.

This is obviously nothing that directly improves the user experience, so it is a bit of an unthankful but important task. At least for new tests, we should strive to make use of pytest's features and update the existing tests by and by.

[kba]

See https://github.com/OCR-D/core/pull/581/files#diff-68b95183e452cd1438e45abf6eba00d7 on how a test looks like that makes use of pytest/unittest features.