Make Metafunc.parametrize keyword-only from indirect onwards

[Zac-HD]

Passing indirect/ids/scope positionally was a common source of confusing errors -- most notably parametrize("arg1", "arg2", [(1, 2)]) sending the argvalues into indirect and failing with "indirect fixture '(1, 2)' doesn't exist" (closes #8593).

The type annotations for the pytest.mark.parametrize stub have declared these arguments keyword-only all along; this aligns the runtime signature.

Technically a breaking change, but I think the overwhelmingly common case is that it was a confusing error, not a deliberate positional use.

[RonnyPfannschmidt]

might want to do a grep.app search as community service - there mgiht be missed breaks for downstreams

[Zac-HD]

I had Claude run some scans of the ecosystem:

1. PyPI top-5000 AST scan. Downloaded the sdist of each of the top 5000 packages by downloads. 4,905 packages scanned successfully (564,808 Python files), containing 135,419 parametrize calls. Zero would break.

2. GitHub-wide AST sample. Via Sourcegraph's search API, fetched 467 complete files across 140 repos matching pytest.mark.parametrize( and ran the same AST analysis: 1,439 calls, zero breaking.

3. GitHub-wide regex counts. Against a baseline of ~1.57M pytest.mark.parametrize( matches on Sourcegraph, precision-first regex patterns found: positional bool 4–9, string 3, list 2–3, identifier ≤9 — 10–25 affected call sites, of which half are errors.

Given the benefits to new users, I'm comfortable merging.