Handle VLA Q descriptor in TableDataDiff

[casey-brooks]

Related Issue

• Fixes FITSDiff reports differences for identical files with Q VLA columns #107

Summary

• treat Q variable-length array descriptors exactly like the existing P descriptors in TableDataDiff
• add regression coverage for QD/QJ VLA columns, mixed P/Q tables, and printdiff text for identical files

Reproduction (before fix)

1. Create a binary table containing a QD() variable-length array column and write it to disk:

   from astropy.io import fits
   from astropy.io.fits import Column
   import numpy as np
   
   data = np.array([
       np.array([1.0, 2.5]),
       np.array([], dtype=float),
       np.array([3.3]),
   ], dtype=object)
   
   fits.BinTableHDU.from_columns([Column(name="a", format="QD()", array=data)]) \
       .writeto("q.fits", overwrite=True)

2. Compare the file to itself:

   from astropy.io.fits.diff import FITSDiff
   
   diff = FITSDiff("q.fits", "q.fits")
   print(diff.report())

3. The same output appears when running fits.printdiff("q.fits", "q.fits").

Observed output (diff report only, no stack trace):

 fitsdiff: 0.0.0
 a: q.fits
 b: q.fits
 ...
   Data contains differences:
     Column a data differs in row 0:
        at [0]:
          a> [1.0 2.5]
          b> [1.0 2.5]
     1 different table data element(s) found (33.33% different).

Testing

• pytest astropy/io/fits/tests/test_diff.py -q
• pytest astropy/io/fits/tests/test_diff.py -k vla -q

[casey-brooks]

Local Testing

• pytest astropy/io/fits/tests/test_diff.py -q (53 passed)
• pytest astropy/io/fits/tests/test_diff.py -k vla -q (5 passed, 48 deselected)

[noa-lucent]

Looks good; Q VLAs now follow P path and tests cover key scenarios.