R4.4: export and copy read as the visual mode reads

HTML export, plain text and rich copy read each paragraph, heading,
cell and HTML block through the component's MarkdownReader (pin
1251221). Emoji, alerts, the TOC and inline tags match visual mode.
Footnotes sit in place as [n]; link titles are not written.

Three review rounds failed; round 3's one finding (a scan per
footnote over open marks) is fixed with a growth regression, unreviewed.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

scripts/fetch-references.sh

@@ -36,7 +36,7 @@ fetch() { # name url rev [sparse paths...]
   echo "fetch $name $rev"
 }
 
-fetch Ely-GPUI-Components https://git.zacharyzhang.com/ZacharyZhang-NY/Ely-GPUI-Components.git 9ff403df31894b07904cfd197a074d1ff95e6de1
+fetch Ely-GPUI-Components https://git.zacharyzhang.com/ZacharyZhang-NY/Ely-GPUI-Components.git 12512217bd9bc652ed6bbf10461b00cc26c01c21
 fetch zed https://github.com/zed-industries/zed.git 1a28cff4b409169bac058bca40dfbfeb7621d19b
 fetch libreoffice-core https://github.com/LibreOffice/core.git 8221e31b3ac356a1623c672912a3d2b492f7e3d1 \
   include/LibreOfficeKit libreofficekit desktop/source/lib desktop/inc \