fix(muya): inline export CSS + heading ids for offline export & TOC anchors (PG7/PG8) - #4412

… heading-link, source cursor, saved indicator) (#4415)

* fix(desktop): map copyAsRich to the engine copyAsRich method (PG9)

The legacy "Copy as Rich Text" command was remapped to copyAsHtml, which
blanks text/html and puts the HTML source into text/plain, so pasting into
Word/email yielded raw HTML markup as literal text. #4411 added a real
Muya.copyAsRich() that writes rendered HTML to text/html and plain text to
text/plain; point COPY_PASTE_METHOD_MAP.copyAsRich at it.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

* feat(desktop): wire heading-copy-link to copyGithubSlug (PG11)

#4414 made the engine attach a hover-to-copy affordance to every heading and
emit heading-copy-link { key } (key == the heading's stable slug) on click.
Re-subscribe in editor.vue and forward the key to editorStore.copyGithubSlug,
which copies `#<githubSlug>` to the clipboard — restoring the heading-anchor
copy affordance that was a documented gap after the @muyajs/core migration.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

* fix(desktop): match exported-TOC anchors to engine heading ids (PG8)

editor.vue now exports via @muyajs/core (#4406), and #4412 injects
github-compatible heading ids onto exported headings (deduped in document
order with a `-N` suffix). getHtmlToc still slugged via the legacy muyajs
Slugger, so `href="#slug"` targets no longer matched the injected ids and TOC
/ [TOC] links were dead. Swap to @muyajs/core's generateGithubSlug and
replicate the engine's whole-document `-N` dedup so the anchors resolve.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

* fix(desktop): consume selection-change affiliation for menu state (PG1)

adaptSelectionChange hardcoded affiliation:[] and copied changes.type
('Caret'/'Range') into start.type/end.type, so createApplicationMenuState's
`start.type === 'span'` and functionType guards never fired — the native
Paragraph-menu check marks, loose/task-list toggles, table/code-fence
detection, and Format-disable-in-code all went dead after the @muyajs/core
migration.

#4410 added an `affiliation` chain (outermost-first) plus per-endpoint
`anchorBlockInfo`/`focusBlockInfo` (`type: 'span'` + `functionType`) to the
selection-change payload. Map them onto the legacy shape:
  - start/end `.type` and `.block.functionType` from the leaf block info,
  - affiliation passed through, with a derived `functionType` surfaced on
    `pre`/`figure` containers so table / code-fence detection lights up.

Also fix the consumer's loose-list read: the engine affiliation entry carries
`isLooseListItem` on the list block directly, not via a `children` chain.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

* fix(desktop): restore saved indicator on undo-to-disk (PG15)

makeSyntheticHistory minted an ever-incrementing editSeq id on every
json-change (including undo/redo), so after edit-then-undo-to-saved the id
never matched lastSavedHistoryId and the tab stayed marked dirty even when its
content matched disk.

Derive the synthetic id from the engine undo-stack DEPTH instead — a stable
position marker that returns to its saved value when an edit is undone back to
the baseline. Seed lastSavedHistoryId to 0 (the engine's post-setContent
baseline depth, since setContent clears history) so a freshly-loaded,
never-saved document clears its dirty indicator when undone back to disk
content, mirroring the legacy history-index behaviour.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

* feat(muya): add setCursorByOffset for source-mode cursor restore (PG2)

The source-code -> WYSIWYG handoff carries only a CodeMirror {line, ch} index
cursor; @muyajs/core had no index-offset -> block-path conversion, so the
WYSIWYG caret was lost. Add Muya#setCursorByOffset, reproducing the legacy
muyajs approach: inject sentinel strings into the current markdown at the
line/ch offsets, rebuild the tree (sentinels embed as literal text), find the
content blocks they landed in, then rebuild the clean document and set the
cursor by the resolved block paths + offsets. Both setContent calls run
synchronously so no intermediate paint occurs, and the method is a no-op for
stale/unresolvable cursors.

Engine helper only — desktop wiring lands in a separate commit.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

* fix(desktop): restore WYSIWYG caret after source-mode edit (PG2)

handleFileChange dropped the saved muyaIndexCursor on the source-code ->
WYSIWYG handoff, so the caret was lost. Consume the new engine
Muya#setCursorByOffset: when the tab has no key-based cursor but carries a
CodeMirror {line, ch} index cursor, map it onto a block-key cursor so the
caret lands where the source-mode cursor was. Restore the per-tab engine
history afterwards (setCursorByOffset re-runs setContent internally, which
clears history).

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

* test(desktop): flip parity e2e for PG1/PG2/PG15, defer PG14

Remove the `test.fail()` markers on the PG1 (Paragraph-menu check mark), PG2
(source-mode caret restore), and PG15 (saved indicator on undo-to-disk) parity
e2e tests now that the desktop wiring lands — all three genuinely pass.

PG14 (first undo after source mode reverts the bulk edit in one step) stays
`test.fail()`: recording the source-mode change as a single undo boundary needs
a general whole-document json1 diff through Editor.updateContents' pick/drop
walker, which only handles specific op shapes and would risk corrupting the
document. Deferred with an explanatory note here and in handleFileChange.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

* docs(test): reconcile parity scoreboard after wave-2 (14/15 fixed)

Update the Status column to the true post-merge state: all seven Wave-1 engine
PRs (#4408-#4414) plus the Wave-2 desktop wiring close PG1-PG13 and PG15. Fix
the "Gaps remaining" count (11 -> 1), add the PG2 setCursorByOffset engine spec,
and document why PG14 is accept-deferred (single-undo-boundary across the
source-mode handoff needs a whole-document json1 diff the op walker can't safely
apply).

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

* fix: address Copilot review on parity wave-2

- editor.vue adaptSelectionChange: restore start/end block.text from the live
  anchorBlock/focusBlock so SELECTION_CHANGE can still slice the selected text
  (search prefill); the previous {functionType}-only block dropped it.
- editor.vue isIndexCursor: validate both line AND ch are numbers (factored out
  isIndexPosition) so a missing ch no longer silently clamps to column 0.
- muya setCursorByOffset: snapshot getHistory() and restore it after the
  internal setContent rebuild so the public API is caret-only and does not clear
  the undo stack; document the behaviour and cover it with a test.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

---------

Co-authored-by: Claude Opus 4.8 (1M context) <[email protected]>