[endrebak]

Preserve chromosome names as text when pandas 3.1 reads BED, GFF3, peak and signal files: requesting categorical dtypes directly can infer integers and merge distinct names such as "01" and "1". Categorize after parsing, retain pandas/Arrow parity, and scope the no-pyarrow fixture to PyRanges1 imports so it does not invalidate pandas' cached optional-dependency state.

Require ruranges >= 0.2.8 for contiguous normalization at the Rust boundary and gtfreader >= 0.3.1 for the equivalent GTF reader fix. Add 13 strided-column integration cases with coordinates beyond int32 and make the column-display doctests insensitive to whitespace. Bumps PyRanges1 to 1.5.1.

Prerequisites: merge and publish these backend fixes before merging/releasing this dependency update:

• ruranges: Normalize strided inputs before calling Rust kernels ruranges_py#42
• gtfreader: Preserve textual GTF categories with pandas 3.1 gtfreader#4

Validation with Python 3.12 / pandas 3.1.0 and rebuilt backend wheels:

• Ruff formatting and lint checks pass.
• ty 0.0.87 check passes.
• pytest --doctest-modules pyranges1/ tests/unit/: 969 passed, 9 skipped.
• Additional BED/GFF3/narrowPeak/bedGraph checks preserve distinct "1" and "01" chromosome names on the applicable pandas/Arrow paths.

PR CI builds wheels from pinned companion backend commits and supplies them to tox through PIP_FIND_LINKS, allowing validation before PyPI publication. Local full tox py312 run passes, including tutorials, package doctests, 860 unit tests, typing and Ruff.