Dynamically added xfail mark does not work on fixtures preparations phase

Normally, decoration with xfail prevents from failure due to ERROR's in loading fixtures. But sometimes decorations are not possible as they cannot load functions or lazy-load other fixtures or configoptions. See also https://github.com/pytest-dev/pytest/issues/7395

Dynamically applied marks should do the same e.i. resolves to XFAIL on running pytest versus `test_xfail instead of ERROR.

@pytest.fixture
def conditional_fxt():
    return "a" 

@pytest.fixture
def mark_xfail_conditionally(request, conditional_fxt):
    if conditional_fxt == "a":
        request.node.add_marker(pytest.mark.xfail(reason="test"))

@pytest.fixture
def setup(mark_xfail_conditionally):
    raise RuntimeError()

def test_xfail(setup):
    ...

also didn't not work with request.applymarker (IDK the usecases BTW).

Workarounds that are not always possible/desirable:

• using decoration
• using pytest.xfail instead of marks (won't run the further code)
• making setup fixture callable

@pytest.fixture
def setup(mark_xfail_conditionally):
    def fn():
        raise RuntimeError()
    return fn

def test_xfail(setup):
    setup()
    ...

Checked on:

• pytest 6.2.2, pytest 7.2.2
• ubuntu

Start with the fixture preparation phase and the handling of dynamically added xfail marks, using the provided conditional_fxt, mark_xfail_conditionally, setup, and test_xfail reproduction. Compare dynamically applied marks with decoration and request.applymarker behavior. Done means a fixture-preparation RuntimeError is reported as XFAIL rather than ERROR when the xfail mark is added dynamically.