M4.86: PDF page thumbnails and fast web view (linearised files)

Embed Page Thumbnails gives every page a /Thumb image: the page drawn on
white, 106 px on its long side, without the layers the page leaves out
(the engine draws them; vectorcraft_pdf::page_areas says what each page
shows; PDF/X files get grey ones). Optimize for Fast Web View writes a
linearised file: the objects are put in linearised order and numbering,
encrypted when a password asks (the encryption dictionary goes with the
catalog, PDF layers' groups too), then laid out with the linearization
dictionary, both cross-reference tables and the page offset and shared
object hint tables (the hint stream encrypted like any other). Their
"not applied yet" warnings are gone. Reading the objects back uses the
lexer of encrypt.rs (syntax.rs adds what rewriting needs).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

docs/mcp.md

@@ -231,6 +231,9 @@ Images follow the `compression` settings of their kind (`color`, `gray`, or `mon
 with `zip`, `jpeg` (at `quality`; images with transparency stay lossless) or `auto` (JPEGs stay JPEG, the others
 lossless). `none`, `jpeg2000`, CCITT and `runLength` are written as ZIP, with a warning when an image needs them.
 `document.pdfSettings` lists the options that differ from the preset and the warnings without writing a file.
+`thumbnails: true` embeds each page drawn small (106 px on its long side, without the layers the page leaves out) as
+its `/Thumb` image. `fastWebView: true` writes a linearised file (the linearization dictionary first, then the first
+page with hint tables saying where every other page's objects are), which stays linearised when a password encrypts it.
 
 Opening a PDF (or `.ai`) imports every page as an artboard and layer; `document.open` takes `pages` ("2-3, 5", 1-based),
 `cropTo` (`bounding` (the art's bounds), `art`, `crop` (default), `trim`, `bleed`, `media`: the box each artboard gets)