My attempt to fix #136601.

• This should be merged after gh-136602: Fix wrong offset_data being used for sparse files with an additional size PAX keyword #136621 because it would make my case even worse, because the TAR offset recomputation would use the real file size instead of the TAR data size.
• Should I add the minimal reproducer from Next TAR header offset recomputation is wrong for GNU sparse 1.0 file combined with 'size' PAX header key #136602 into Lib/test/archivetestdata/ and add a test into Lib/test/test_tarfile.py? The test setup seems a bit complicated. I'm unsure whether I would do it the right way. That reproducer would be a test case for both bugs / pull requests.

• Issue: tarfile indeterminate TarInfo.size when PAX headers contain size and GNU.sparse.realsize keys at the same time #136601