TST: Change class-based auto-use setup to fixtures

[TomAugspurger]

In #19535 we're excluding data test files from the distribution, which required some changes.

The hardest tests to update were the IO ones that read in a bunch of files in setup and set them as attributes on self.

In the PR, I changed these setup / teardown methods to be yield-base autouse yield methods. See https://github.com/pandas-dev/pandas/pull/19535/files#diff-8cfd1704100c1e13de1bac288482b344R41 for an example.

Ideally we would change the tests to accept the file / parsed object as a fixture, rather than accessing it off self.

The hard part is generating fixtures dynamically, e.g. one fixture per CSV in this directory. We can accomplish that with pytest_generate_tests: https://github.com/pandas-dev/pandas/pull/19535/files#diff-3f782eafb799811bd5516f328411418dR31

[TomAugspurger]

In #19535, we decided that auto-generating the fixtures with pytest_generate_tests is perhaps a bit too magical. Wherever possible, it'd be better to just explicitly write out the fixtures, e.g. like

@pytest.fixture(params=[
    'chinese_utf-16.html',
    'chinese_utf-32.html',
    'chinese_utf-8.html',
    'letz_latin1.html',
])
def html_encoding_file(request, datapath):
    """Parametrized fixture for HTML encoding test filenames."""
    return datapath('io', 'data', 'html_encoding', request.param)

[mroeschke]

I don't seem many old style def setup_* methods anymore in the test suite (especially related to IO functionality), so I think we can close this issue