Annotate the pmcid arrays as ndarray, not Sequence[int]

[adelavega]

make test_mypy currently fails on main:

src/pubget/_vectorization.py:200: error: Incompatible return value type
  (got "tuple[ndarray[tuple[Any, ...], dtype[Any]], dict[str, Any], Any]",
   expected "tuple[Sequence[int], dict[str, Any], Any]")  [return-value]

_extract_word_counts returns np.concatenate(...), an ndarray, while its
signature promises Sequence[int]. Older numpy stubs allowed an ndarray to
satisfy Sequence[int]; from numpy 2.1 the shape-parameterised
ndarray[tuple[Any, ...], dtype[Any]] no longer does. That is why the failure
appeared without any change to this file — the py313-latest-deps leg of the
matrix now resolves a numpy new enough to reject it, while py310-min-deps
still passes.

_vectorize_articles has the same mistake one level down: it returns
articles["pmcid"].values under the same Sequence[int] annotation. mypy cannot
currently see it, because DataFrame.values is Any without pandas stubs, but it
would surface as soon as those are installed. Both are fixed here.

Annotations only — no behaviour change.

Verification

Reproduced the CI failure locally in a clean venv with numpy 2.5.3 and mypy 2.3.1,
then confirmed the fix under both extremes:

numpy	before	after
2.5.3	the error above	Success: no issues found
1.25.2	clean	clean

The module still imports and the annotation resolves to a real numpy.ndarray
type at runtime.

🤖 Generated with Claude Code