chore(deps): move the tiptap stack to peerDependencies, following upstream — next major

[IgorShevchik]

Background

Split out of #99 §4 while doing §3 (#351). What made it worth its own issue is that the situation changed since #99 was filed: upstream has already done this, so it is now a divergence question as well as an install-weight one.

Where we are

package.json dependencies currently holds 64 runtime entries, among them:

group	count
@tiptap/*	17
embla-carousel-*	7
@tanstack/*	2

Every consumer installs all of it. An app that uses B24Button and B24Modal and never touches B24Editor still pays for the full ProseMirror stack.

What upstream did

nuxt/ui@v4 now declares 21 entries in peerDependencies that we carry as dependencies — the whole tiptap set plus ai:

"@tiptap/core": "^3",
"@tiptap/extension-bubble-menu": "^3",
…
"@tiptap/vue-3": "^3",
"ai": "^6 || ^7"

Of those, only ai is marked optional in peerDependenciesMeta; the tiptap entries are required peers upstream. Worth confirming that is intentional on their side before copying it wholesale — required peers for an editor most apps do not use would be an odd choice, and peerDependenciesMeta: { optional: true } is the shape that actually matches "install it if you use the Editor".

The practical consequence for us: every future tiptap port now has to reconcile two different dependency shapes. That cost grows.

Why this is not a patch

Moving a package from dependencies to peerDependencies is breaking for consumers — anyone using B24Editor must add 17 packages to their own manifest or their build stops resolving. It belongs in the next major, with a migration note, and embla-* deserves the same question asked separately (the carousel is a smaller stack and a more commonly used component).

Suggested shape

1. Measure first: install size and node_modules count for a minimal app today versus with the tiptap stack removed. chore(deps): adopt Renovate; declare vue in peerDependencies; document exact-pin rationale (reka-ui, vaul-vue, h3, unimport); review dependency weight #99 asked for this and it has never been done — without it we are trading a real breaking change for an unquantified win.
2. Decide required-vs-optional peers deliberately rather than by copying upstream. optional: true plus a clear runtime error from B24Editor when the peer is missing is friendlier than a resolution failure at build time.
3. Ship in the next major with a migration snippet for package.json.
4. Record the outcome in .sync/PORTING.md §2 either way — if we keep them as dependencies, that is a deliberate divergence a port would otherwise "fix".

Not in scope

@tanstack/*, fuse.js, canvas-confetti, motion-v — small enough that the breaking change would cost more than the weight saved. Revisit only if the measurement in step 1 says otherwise.