2.0: a pass that reads latest(s) before s() computes twice on a plain write #4402

On next @ 7ebd4a3 (and on #4401), a reader whose compute reads latest(s) before s() computes twice on a single plain synchronous write (no action, no async). Reading s() first computes once (pinned in #4401).

const [s, set] = createSignal(0);
createRenderEffect(() => [latest(s), s()], () => {}); // compute runs twice per set()
set(1); flush();

The same happens inline in JSX or through a memo, and with a memo, render effect or user effect as the reader.

Why it happens. A pass that has read only latest() so far enters the verdict lane on purpose. That is what lets a Show over latest() mount its child once (#3540, boundary-not-born-held-3540 pins mounts [1, 2, 2]). The plain s() read that follows is then lane work, so it is served the screen (D-37) and runs again at the landing.

What was tried in #4401. Neither attempt worked:

• Keeping latest-only passes in the frame's seat makes Loading on should behave as a keyed <Show> around the boundary; a boundary mounted under a hold must not be born held (A29) #3540 mount 3 times.
• Letting the verdict lane's seam re-derive a staged read only if the flush parks breaks the inline latest-first park pin (v:0 l:1 beside the held write).

A fix needs the verdict lane's seam to treat verdict readers differently from lane work. The rules involved are D-22 (one pass; re-derive once only if the flush parks), D-37, F-10 and the #3540 ruling.

Expected: one compute per plain write whatever the read order, as for [s(), latest(s)].