fix: swev-id: astropy__astropy-14539 handle FITSDiff Q VLAs

[casey-brooks]

Issue

Closes #127.

Observed failure

• FITSDiff skipped Q-format variable-length array (VLA) columns because the comparison logic only treated P descriptors as VLA data, so row-wise differences went unreported.

Reproduction

1. Create two binary tables that include a format="QE" column.
2. Modify the VLA payload in one row of the second table.
3. Run FITSDiff(file_a, file_b, rtol=0, atol=0) — .identical incorrectly returns True and the report omits the column.

Changes

• Update TableDataDiff to treat Q-format descriptors as VLA columns alongside P descriptors.
• Add regression tests covering Q-format VLAs (identical self-compare, differing payload, mixed fixed-width columns) plus a parity check for the existing P-format path.

Testing

• PYTHONPATH=. LD_LIBRARY_PATH='/nix/store/sbfrl6127w424l34lsq0lzryibq1226j-gcc-15.2.0/lib:/nix/store/h4qhxh7vwmxgy6w05g0xsf6r1bfi9vga-gcc-15.2.0-lib/lib:/nix/store/dvb9i3s69j29md1hpx3gpaqqds6530df-gcc-15.2.0-libgcc/lib:/nix/store/n5lymg0y5x6i9wipkjrsi8aczv1nr4qc-zlib-1.3.1/lib' SETUPTOOLS_USE_DISTUTILS=stdlib .venv/bin/python -m pytest astropy/io/fits/tests/test_diff.py -k vla_q (3 passed)
• PYTHONPATH=. LD_LIBRARY_PATH='/nix/store/sbfrl6127w424l34lsq0lzryibq1226j-gcc-15.2.0/lib:/nix/store/h4qhxh7vwmxgy6w05g0xsf6r1bfi9vga-gcc-15.2.0-lib/lib:/nix/store/dvb9i3s69j29md1hpx3gpaqqds6530df-gcc-15.2.0-libgcc/lib:/nix/store/n5lymg0y5x6i9wipkjrsi8aczv1nr4qc-zlib-1.3.1/lib' SETUPTOOLS_USE_DISTUTILS=stdlib .venv/bin/python -m pytest astropy/io/fits/tests/test_diff.py (52 passed)
• SETUPTOOLS_USE_DISTUTILS=stdlib .venv/bin/tox -e codestyle -- --files astropy/io/fits/diff.py astropy/io/fits/tests/test_diff.py (pass)

[casey-brooks]

Local test & lint results:

• PYTHONPATH=. LD_LIBRARY_PATH='/nix/store/sbfrl6127w424l34lsq0lzryibq1226j-gcc-15.2.0/lib:/nix/store/h4qhxh7vwmxgy6w05g0xsf6r1bfi9vga-gcc-15.2.0-lib/lib:/nix/store/dvb9i3s69j29md1hpx3gpaqqds6530df-gcc-15.2.0-libgcc/lib:/nix/store/n5lymg0y5x6i9wipkjrsi8aczv1nr4qc-zlib-1.3.1/lib' SETUPTOOLS_USE_DISTUTILS=stdlib .venv/bin/python -m pytest astropy/io/fits/tests/test_diff.py -k vla_q → 3 passed
• PYTHONPATH=. LD_LIBRARY_PATH='/nix/store/sbfrl6127w424l34lsq0lzryibq1226j-gcc-15.2.0/lib:/nix/store/h4qhxh7vwmxgy6w05g0xsf6r1bfi9vga-gcc-15.2.0-lib/lib:/nix/store/dvb9i3s69j29md1hpx3gpaqqds6530df-gcc-15.2.0-libgcc/lib:/nix/store/n5lymg0y5x6i9wipkjrsi8aczv1nr4qc-zlib-1.3.1/lib' SETUPTOOLS_USE_DISTUTILS=stdlib .venv/bin/python -m pytest astropy/io/fits/tests/test_diff.py → 52 passed
• SETUPTOOLS_USE_DISTUTILS=stdlib .venv/bin/tox -e codestyle -- --files astropy/io/fits/diff.py astropy/io/fits/tests/test_diff.py → lint passed

[noa-lucent]

Looks good. Treating Q-format descriptors as VLAs in TableDataDiff resolves the false positives, and the new tests cover self-compare, delta-detect, mixed columns, and parity with P formats.