Fix #861: open .ai files whose non-native art has ASCII85 lines starting with %_

[rschuerer]

Fixes #861.

A .ai saved by Illustrator 28 without PDF compatibility wouldn't open: "this file was saved without PDF compatibility, and its editing data can't be read (it ends inside a layer)".

Commit 1: open the file

Its layer starts with a /ForeignObject (non-native art: the content of a placed PDF), whose /Data entry is that PDF as ASCII85 in % comment lines. ASCII85 includes _, so about one line in 85 starts with %_, which the lexer reads as hidden tokens; a ( in such a line opened a string that swallowed the dictionary's ;, the rest of the art and the layer's LB. The reader now reads that data as a whole, up to its ~>, as it already did for a /Binary dictionary's ASCII85.

Commit 2: draw the non-native art

The letterhead's main graphic, a fan of swooshes, is exactly that object: a radial gradient in a 3-point PDF, scaled 271 times and clipped by the swoosh paths. Leaving it out opened a page without its graphic.

The reader now decodes the PDF, has the PDF importer read it, and fits its art to the object's box as a group named "Non-native art". The box is /Bounds measured from /Origin with y down, then /RTransform; on both objects of that file it lands exactly on the clipping paths around them. The PDF's content keeps its own y-up coordinates, so it is flipped into the box (verified against the original: the gradient fades from the apex at the top right). Ids are the document's, the PDF's swatches stay unlinked, its images join by content key. A PDF that can't be read is left out with a note.

The EPS crate now uses the PDF crate, both layer 3; xtask layers allows it like render and metafile.

Tests: a data line starting with %_( that isn't a PDF (opens, note), and a real tiny PDF with two squares fitted to the box, checking placement, orientation, colours and ids. docs/mcp.md updated. cargo xtask ci passes.

Not in this PR: showing a failed open's error in the window (noted in #861).

🤖 Generated with Claude Code

https://claude.ai/code/session_017ZwZjADUZP4VzvrFGpiZch

[bflatastic]

Thanks again, @rschuerer! This is a neat find: an ASCII85 line that happens to start with %_ turning into hidden tokens. Skipping to ~> after /Data , the same way as /Binary is the right fix, and leaving the non-native art out with a note is honest. It's in my CI queue now; I'll merge it once the full cargo xtask ci passes here.

[bflatastic]

Merged, thanks @rschuerer! Tracking a file that "ends inside a layer" down to an ASCII85 line that happens to start with %_ is a lovely piece of detective work. Skipping to ~> after a /ForeignObject's /Data ,, the same way the reader already does for /Binary, fixes it at the root, and leaving the non-native art out with a warning (rather than silently) is the honest choice until we can draw it. The test pins exactly the bad case, a ( inside a %_ line.

I rebased it onto current main (only docs/mcp.md had moved, no conflict); the full cargo xtask ci passed on that head, all 7 steps.