v1.1.5

1.1.2 to 1.1.4 were @matrajs/mcp on its own, carrying regenerated
documentation to the MCP registry. Publishing the rest at 1.1.2 would have left
core behind mcp and made the line at the top of this file false, so the whole
set moves to 1.1.5 and shares a number again.

renderToHTML(doc, extensions) · a stored document to HTML with no editor.
Generic rather than a switch over the built-ins: every element comes from the
extension's own toDOM, the same function the editor draws with, so an
extension written this afternoon renders on a server without touching core and
the two cannot disagree about what a callout looks like. createEditor(). getHTML() already worked without a DOM; this skips the schema build as well and
is 2.8× faster over 500 documents. A parity suite renders eight documents both
ways and compares them byte for byte.

contentElement on a parse rule · where a node's content actually is. A
selector, or a function returning the element holding it. Without it the matched
element's own children are the content, which is wrong for any tag that wraps
its content beside something else.

Two new nodes. audio is a block with the source checked the way an image's
is. ruby annotates CJK text — the base is editable content and the reading is
an attribute, which is only possible because of contentElement: <ruby> holds
both, so parsing all of it puts the reading inside the word it annotates.

A node decoration lands on the node its range spans, and nowhere else. The
renderer applied a node decoration to every element its range overlapped, and
a range always overlaps every ancestor of its own node. The visible version:
placeholder({ everyBlock: true }) on an empty list paragraph put its prompt
on the paragraph, the item and the list, three ghosts stacked over real
content, and the caret looked lost underneath them. An extension that wants
the ancestors marked says so with one spec per ancestor, which is what
focus({ ancestors: true }) has done all along; the renderer now agrees with
it, in the build path and the patch path both.

A trailing space is visible the moment it is typed. The document kept the
space and the screen did not show it until the next character landed, because
default rendering collapses trailing whitespace and the view never said
otherwise. It now sets white-space: pre-wrap (and overflow-wrap: break-word) on the editable element, only when the host has not chosen a
white-space of its own; the check reads computed style, so a stylesheet choice
counts as choosing. The opening comment of whitespace.test.ts had described
this exact symptom while testing only the document half of it.