Core: byte-swap the float vector data, not the attribute struct (fixes big-endian crash)

[Scottcjn]

Addresses the testAttributes failure in #1175.

extract_attr_float_vector reads the float array into attrdata->arr, then byte-swaps with priv_to_native32(attrdata, n). On big-endian hosts that swaps the first n 32-bit words of the exr_attr_float_vector_t struct (the arr pointer and length) instead of the n floats the array holds. The corrupted pointer and length then crash when the attribute is read back: OpenEXR.testAttributes segfaults constructing the FloatVector in Context::header (a std::vector<float> built from a garbage pointer range with a ~64M element count).

The fix swaps attrdata->arr instead. priv_to_native32 is a no-op on little-endian, so x86_64 and ppc64le are unaffected.

The swap runs only inside if (rv == EXR_ERR_SUCCESS && n > 0), after exr_attr_float_vector_init has allocated attrdata->arr and sequential_read has filled it, so attrdata->arr is always valid and non-null at this point.

Testing

On real POWER8 hardware (RelWithDebInfo build, crash localized with gdb):

• ppc64 big-endian (gcc 13.3): OpenEXR.testAttributes segfaults before this change and passes after.
• Little-endian (x86_64, ppc64le) is unchanged: the swapped call does nothing on little-endian, so any EXR with a float-vector attribute reads identically there.

[Scottcjn]

Quick heads-up: the unstable state here is not a CI failure. The GitHub Actions workflows are sitting at action_required (awaiting maintainer approval to run), not red. The sibling PR #2509 had its workflows approved and passed clean. If a maintainer can approve the run here and take a look, this should green up the same way. The change is the one-line byte-swap-target fix in extract_attr_float_vector (swap the float array, not the attribute struct header), validated on a native big-endian ppc64 host where testAttributes segfaults before the fix and passes after.

[cary-ilm]

Thanks! I poked the CI and everything passes. It looks good to me but I'd like @kdt3rd to look at it before merging. Thanks for the fix. I remember that issue coming in ages ago but wasn't sure what to do about it, so we really appreciate the help.

[Scottcjn]

Thanks for the opportunity to use my weird knowledgebase and skillsets to help ilm.
😆

[kdt3rd]

lgtm - thanks for noticing that!