Preserve single-pass iterables in stack and concat

[sylvesterkaczmarek]

Materialize the input iterable in stack and concat before creating their field-processing callbacks. Currently _ops_base materializes a generator locally, but the callbacks retain the original, exhausted iterator. Both operations then fail while stacking or concatenating the first array field.

The two-line fix shares one materialized list across all array and nested-dataclass fields. List-input behavior, static metadata, axis handling, and backend selection are unchanged.

Validation

macOS arm64 CPU, Python 3.12.11; NumPy 2.5.3, JAX 0.11.2, TensorFlow 2.21.0, PyTorch 2.14.0:

• All 40 new regression cases fail on unchanged main and pass with the fix.
• Coverage includes generators, list iterators, a custom single-pass iterable, multiple fields, nested dataclasses, both operations, and two batch axes on all four backends.
• Full package suite: 425 passed (python -m pytest -q dataclass_array --tb=short).
• Built and installed the wheel outside the checkout; all 40 new cases pass against the installed package.
• New-test Pyink formatting, Python compilation, and git diff --check pass. Ruff E9/F passes with F722 excluded for the project's shape-annotation syntax.

Accelerators and documentation builds were not run. This does not change negative-axis handling and is independent of #104.