Fix load_npz single-variable skip over extra fields

[realwangzhichao]

Summary

This PR fixes xt::load_npz(filename, search_varname) when scanning NPZ files that contain a non-zero ZIP local header extra_field_len.

Previously, when the current array name did not match search_varname, the code skipped only entry.compressed_size. This left the stream positioned inside the current entry's extra field for NPZ files with extra_field_len > 0, causing subsequent local headers to be misread and later arrays to be reported as missing.

The fix reuses extract_zip64_compressed_size(stream, entry), which already consumes the extra field and handles ZIP64 size metadata, then seeks over the compressed payload.

Validation

I tested this change with a NumPy-generated .npz file whose local ZIP entries each have extra_field_len = 20.

Before this fix, reading an array that appears after the first entry failed:

xt::load_npz<float>(filename, "body_quat_w");

[JohanMabille]

Thanks for the fix. The CI needs to be fixed before we can merge this one, I'm working on it.

[JohanMabille]

Thanks again!