gamut-dng: an interior main-chain page with no image data reaches no decode surface

[justin13888]

Found while closing a review finding on #442 (PR #508). That PR fixed the last main-chain
directory; this issue is the residue, which predates it.

The gap

DngDecoder::decode surfaces every unmodelled IFD field through four verbatim channels:

• DecodedDng::ifd0_extra — IFD 0,
• DecodedDng::raw_extra — the raw sub-IFD,
• SubImage::extra_tags — each directory that decodes as an image,
• DecodedDng::trailing_extra — the last main-chain directory when it is none of those
  (added in feat(dng)!: type the C2PA manifest store and report both exclusion ranges #508, because C2PA 2.4 §A.3.6 lets a manifest store be "the only entity within a new
  IFD following the existing one", which is a directory with no image).

A main-chain page that is neither the first nor the last and carries no image data falls
through all four: decode_sub_image returns None for it, so no SubImage is created and no
extra_tags are attached, and it is not IFD 0, the raw IFD, or the last page. Its fields reach
no surface.

That contradicts the decoder's stated promise, on ifd0_extra: "Nothing those four reach is
silently dropped" — the qualifier is honest, but the underlying intent (issue #109's "all
metadata explicitly represented") is not met for this shape.

Byte accounting is unaffected: deconstruct walks the whole chain and still classifies those
bytes, so the file is fully accounted; it is the typed decode surface that drops them.

Reproducing

Encode any DNG, then append two trailing main-chain IFDs carrying only non-image tags (the
helper append_raw_trailing_ifd in crates/gamut-dng/tests/c2pa.rs builds one at the byte
level). The last page's fields arrive in trailing_extra; the middle page's arrive nowhere.

Options

1. Generalise trailing_extra to every uncovered main-chain page. That means keeping the set
   of main-chain indices from walk_ifds (it currently returns only the last one) and reporting
   per-page rather than as one flat list, since a caller cannot otherwise tell which page a tag
   came from. Renaming the field would be semver-major for gamut-dng.
2. Add a separate per-page channel and leave trailing_extra as the §A.3.6 case it was
   introduced for.

Option 1 is the honest shape; the naming is what makes it more than a one-liner. Either way the
promise on ifd0_extra and the note in gamut-dng/STATUS.md should be updated together with the
fix.

Related

#442, PR #508, epic #109's explicitness clause.