Settle the layout before painting it, not after

Every change the client made painted one frame of the old layout under
the new contents, and one screen never stopped moving at all. Three
separate reasons, all here rather than in the client.

A container is as big as what its children asked for LAST frame, so the
first walk after the reconciler patches the tree knows the new children
and the old sizes. `frame!` would have caught that — it settles before
it draws — except that it skips the settling passes while nothing is
moving, which is almost every frame, and that test looks at what the
last walk did. A patch arrives between two walks and is invisible to it
until the walk that paints. So the backend says what it did:
`unsettle!`, from the `:before` hook, which is handed the context now
for that purpose. The frame that would have shown the old layout spends
itself settling instead.

Three settling passes were not enough either. A size travels up the tree
in one pass and back down in the next, so a change deep inside a screen
— the reader's list inside a split inside a page — was still moving on
the third and settled on the fourth, one pass after it was painted.
Eight, and only the frames that are actually moving pay for them.

And `:max-width` on a page was a floor rather than a ceiling: the column
reported whatever its children asked for, which made its width a
function of its own width. One thing too wide to wrap — a URL, a row of
buttons — dragged the column past the cap and took every paragraph on
the page out with it, and where the sums did not come to rest the column
swung between two widths, one per frame, for as long as the screen was
up. The settings screen did exactly that. `:fixed` there, and the width
is the cap's and the window's; the wrapping under it has something that
holds still to wrap against.

Measured by walking frq's own tree headlessly and asking, per frame,
whether what was painted was still moving. Before: one to three such
frames on every screen change, message, banner and panel, and settings
never settling. After: none.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>