v1.4: trace post-fix livery parent ownership

[Datamining00]

Diagnostic derivative of the parser structural audit.

Why

Audit v2 on the real Livery_0343 sample reports unframed_transform_candidate_count = 0, while XANADU / THE NEW WORLD still renders on the character face. This rules out another still-unframed 16-byte transform as the immediate cause.

The remaining defect is therefore narrowed to a different parser class: a transform can be recognized, yet its parent/child ownership may still be wrong when the GroupNode tree is built.

What this branch does

Adds a non-invasive post-recovery ownership trace. For every livery section it records:

• root direct children;
• every implicit_bare_transform_pair parent;
• each implicit parent's direct children;
• group source offsets, transforms, expected/actual child counts;
• descendant shape/group counts and shape source-offset ranges;
• ancestor group sources/offsets.

The trace is installed after the bare-parent fix and Structural Audit v2, so it captures the tree that is actually flattened and rendered. It does not change decoding, group ownership, transforms, masks, z-order, or rendering.

No car ID, phrase, section-specific coordinate, source offset, or creator is hard-coded.

Expected decision

For the real Right section, the trace will show whether the central XANADU group is actually owned by one of the recovered implicit parents or remains a root sibling. That distinguishes a wrong child-span rule from a different group grammar issue before any additional fix is attempted.

Validation

• Windows CI: success.
• Full regression suite: 178 tests, OK.
• PyInstaller one-file build: success.
• EXE SHA-256: 62762f526e5891514f9c69549fcd6fc41554ab650753c547620c751cb387c8f4.
• Artifact ZIP SHA-256: aa50104ed5749d91d0584def81fe7e008bbd07ad9380e4cf087e74c378319117.

This PR remains draft until the real livery ownership JSON is inspected.