…Slot/ElementDOMSlot split

`triggerCommandListeners` already wraps each listener bucket in
`updateEditorSync`, so `$commitSuggestion` runs in an active editor
context inside the Tab handler. Wrapping in `editor.update({discrete:
true})` from inside that existing update deferred the splice to a
microtask, leaving `didCommit` stuck at `false` long enough for the
EDITOR-priority tab-indentation handler to insert a `<tab>` before the
suggestion landed. Direct `$commitSuggestion()` call instead — the
splice runs synchronously, `didCommit` reflects the real outcome, and
the suggestion-commit case correctly stops propagation.

Touch handler keeps its own `editor.update` wrapper since it is not
called from a command-listener pipeline.

Caught by the rich-text e2e canary on chromium (Autocomplete.spec.mjs
"format-as-the-original-text" and "Undo does not cause an exception")
where the trailing tab created an extra empty <strong> with the
TabNode theme class.

…orator / void cases

The first revision of `prettifyHTML`'s wrapper strip used a non-greedy
regex which terminated at the first `</TAG>` it saw. For an embed-block
wrap (inner is a `<div>` containing nested `<div>` content) the match
stopped at the embed's inner `</div>` instead of the inner-marker's,
so the wrapper scaffolding stayed in the snapshot. Switched to a
depth-aware walker (`findMatchingClose`) and split the strip into three
cases matching the three observed structural shapes:

- ElementNode (paragraph / heading / list etc.) — inner element carries
  the real attributes; merge wrapper attrs (dir / style picked up by
  the reconciler on the keyed DOM) into the inner element, drop the
  wrapper. Wrapper attrs already present on the inner (e.g. an empty
  `style=""` placeholder from indent toggling) are skipped to avoid
  duplicate attributes.
- DecoratorNode with void inner (HR) — the `<hr>` itself is the
  canonical content; merge wrapper attrs into it and emit self-closing.
- DecoratorNode with React portal (embed, PageBreak) — `useDecorators`
  mounts the portal as a sibling of the empty inner-marker since the
  keyed DOM is the wrapper; keep the wrapper minus the marker attr and
  drop the empty inner shell.

`assertSelectionOnPageOrFrame.getPathFromNode` now skips the inner-marker
step when walking up to the contenteditable root, so DOM paths in
selection assertions match the pre-wrap depth.

Tables: `nthTableSelector` switched to Playwright's `:nth-match` since
top-level tables are no longer siblings under wrap. TextFormatting
triple-click test: changed `> p` to descendant `p` so the inner
paragraph (now wrapped) is reachable.

Also fixes a separate Autocomplete commit issue caught at the same
time: the Tab handler wrapped `$commitSuggestion` in
`editor.update({discrete: true})`, but the handler is already inside a
`triggerCommandListeners`-driven `updateEditorSync`, so the nested
update deferred to a microtask. `didCommit` came back as `false`,
propagation continued, and the EDITOR-priority TabIndentation handler
inserted a `<tab>` before the suggestion landed. Direct
`$commitSuggestion()` call instead; touch handler keeps its own
`editor.update` since it isn't called from the command pipeline.

…ctors

The wrap-all approach attempted to give DecoratorNodes (HR, PageBreak,
equation, image) drag handles by wrapping their keyed DOM with a
sibling button + an inner-marker. The latent issue documented in R6/C2
turned out to bite at scale: `useDecorators` mounts the React portal
via `editor.getElementByKey`, which returns the wrapper, so any
decorator whose `decorate()` renders visible content (EquationNode,
ImageNode, etc.) lands the portal as a sibling of the empty inner
shell with undefined visual ordering. Narrow `shouldWrap` to top-level
ElementNode only — DecoratorNodes keep their original keyed DOM and
forgo the drag handle for now. The slot abstraction can cover portal
routing in a follow-up if the wrap-all extension pattern catches on.

Drag handle textContent dropped (`⋮` was a fallback over the
background-image visual). Adding text content to the wrapper interfered
with browser triple-click selection on multi-paragraph documents:
the wrapper became a text-bearing block and the multi-paragraph
triple-click collapsed to an empty selection. The visual is purely the
CSS background-image now, with `aria-label="Drag to reorder"` for
accessibility.

`prettifyHTML`'s wrapper strip generalised for the three structural
shapes observed under wrap:

- ElementNode (paragraph etc.): merge wrapper attrs into inner element,
  drop wrapper. `class` and `style` are composable — wrapper-side
  tokens are appended to the inner's existing value (so e.g.
  alignment-applied `style="text-align: center"` on the wrapper merges
  with the inner's existing class). Other attrs (`dir` etc.) keep the
  inner's value when both define them.
- Void inner (e.g. HR under the legacy wrap, kept for safety): merge
  wrapper attrs into the void element, emit self-closing.
- DecoratorNode wrap (no longer applied today but kept for safety):
  empty inner shell + portal-mounted sibling content. Wrapper-minus-
  marker becomes the result tag.

`assertSelectionOnPageOrFrame`:
- `getPathFromNode` skips the inner-marker layer.
- `normalizeWrapperFocus` translates wrapper-as-focus positions (which
  arrow-key navigation parks at as block boundaries) to the deepest
  first / last text inside the inner so paths line up with the
  un-wrapped baseline.

Spec-level fixes:
- `dragDraggableMenuTo` accepts a `fromBlockSelector` so callers can
  scope the handle lookup to the source block under wrap.
- Direct-child selectors (`div[contenteditable="true"] > p`,
  `.ContentEditable__root > .paragraph`, etc.) switched to descendant.
- `:nth-child` / `:nth-of-type` on per-block elements like
  `tableScrollableWrapper`, `editor-image` switched to Playwright's
  `:nth-match` since the wrap collapses sibling structure.
- `Mutations.spec.mjs` "Remove the paragraph" targets the wrapper
  (the visible top-level block); inner-element manipulations descend
  through `querySelector('p')`.
- `Extensions.spec.mjs` insertText test traverses via
  `editor.querySelectorAll('p')[1]` instead of `firstChild.nextSibling`.

Also revives the Autocomplete commit fix (direct `$commitSuggestion`
without nested `editor.update`) and the unit test for "HR isn't wrapped"
to lock the narrowed `shouldWrap` contract.

Known remaining wrap-systemic regressions (not addressed in this
commit — needs a deeper look at lexical's selection mapping or a
wrap-design revisit):
- Browser arrow-key navigation across wrap boundaries lands selection
  on the wrapper element. `execCommand("insertText")` on a wrap-
  spanning selection drops the inner `<p>`.
- Various text-transform-after-keystroke tests (Hashtags, Mentions,
  Navigation) are affected by cursor positioning across the wrap.

…lize

# Conflicts:
#	packages/lexical-html/flow/LexicalHtml.js.flow
#	packages/lexical/flow/Lexical.js.flow
#	packages/lexical/src/LexicalReconciler.ts
#	packages/lexical/src/nodes/LexicalElementNode.ts

… merge dropped

The merge resolution that resolved conflicts via `git checkout --ours`
inadvertently reverted main changes that the same files needed.

- `LexicalReconciler.ts`: `$reconcileElementTerminatingLineBreak` now
  reads `prevLineBreak` from `slot.element.__lexicalLastChildKind` (cache)
  instead of calling `isLastChildLineBreakOrDecorator(prevElement, ...)`
  on a detached prev-state node — restores the facebook#8548 crash fix.
- `LexicalDOMSlot.ts`: `setManagedLineBreak` writes the cache field that
  the reconciler reads — without this the field declaration was orphan.
- `Lexical.js.flow`: migrates remaining `+T` / `+property` to the new
  `out T` / `readonly` syntax to match facebook#8547's flow migration on main.

…getInsertionAnchor method