Improve ergonomics of image comparison fixture

What kind of feature would you like to request?

Other?

Please describe your wishes

As a continuation of #1772, we should simplify how this is done.

• Decide on fixture api
• Decide how generated outputs are handled (.gitignore-d, deleted each time that test is run, both?)

My proposal for the first TODO is as follows.

In order to test the image files

• ./_images/some_plot/expected.png
• ./_images/some_plot_x_context/expected.png

… against their actual.png counterparts, we currently do

# per module setup
ROOT = HERE / '_images'

# individual test
def test_some_plot(image_comparer):
    save_and_compare_images = partial(image_comparer, ROOT, tol=15)
    ...
    save_and_compare_images('some_plot')
    ...  # and maybe
    save_and_compare_images('some_plot_x_context')

I propose we arrive here (with --strict-markers on):

# per module setup
@pytest.fixture(scope='module')
def compare_images_root():
    """Set image root for save_and_compare_images."""
    return HERE / '_images'

# individual test
@pytest.mark.compare_images_tol(15)
def test_some_plot(save_and_compare_images):
    ...
    save_and_compare_images()
    ...  # and maybe
    save_and_compare_images('x_context')

for that we can make save_and_compare_images

1. use the request fixture to retrieve tol and test name
2. use the test name for the png directory name used.

Start by reviewing #1772 and locating the existing image-comparison fixture and its callers; no implementation files or tests are named. Define the fixture API and generated-output policy, then verify the proposed marker, test-name paths, and strict-marker behavior against the relevant image-comparison tests.