[2.0 / next] An own error over a held write flashes the Errored fallback at the landing when the reader is a memo (inline render effect doesn't) #4407

Summary

A reader computes latest(a) + b() + 1 and throws its own error at one value. a and b are written in the same batch as a slow async memo that holds the write. When the hold lands, the reader's result goes title=2 → [E own:3] → title=6 if it is read through a memo, but title=2 → title=6 if the same function runs directly in a render effect. JSX grouping changes the result (the #4401 / #4284 parity rule: a latest() read inline and through a memo show the same screen).

The content isn't remounted (one mount) and the final screen is right, but the Errored boundary's fallback is shown during the landing flush.

Reproduction

const [a, setA] = createSignal(1);
const [b, setB] = createSignal(0);
const [c, setC] = createSignal(1);
let resolve!: () => void;
createRoot(() => {
  const slow = createMemo(() => {
    const v = c();
    return v > 1 ? new Promise<number>(r => (resolve = () => r(v))) : v;
  });
  createRenderEffect(slow, () => {});
  const f = () => {
    const v = latest(a) + b() + 1;
    if (v === 3) throw new Error("own:3");
    return v;
  };
  const view = createErrorBoundary(
    () => untrack(() => {
      const m = createMemo(f);                  // inline form: createRenderEffect(f, show)
      createRenderEffect(() => m(), v => log(`title=${v}`));
      return "content";
    }),
    e => `[E ${e()?.message}]`
  );
  createRenderEffect(() => view(), v => v !== "content" && log(String(v)));
});
flush();
setA(2); setB(3); setC(2); flush();   // held by `slow`
// ...settle...
resolve();                             // landing

Full probe (memo and store sources, five reader shapes, memo vs effect): zz-own-err-3945.test.ts, run on next at 9baaba3.

Expected

The memo and inline forms show the same sequence, without the fallback in between: the own:3 pass (latest(a) = 2 beside the held b() = 0) isn't a screen the landing should show.

Notes

• Found while reviewing fix(signals): latest() of a memo born in a lane throws NotReady, not undefined (#4362, #4363, #4376) #4383. Its earlier head 4e4091f flashed in both forms; before fix(signals): latest() of a memo born in a lane throws NotReady, not undefined (#4362, #4363, #4376) #4383, next stayed stuck on the error.
• Relevant rules: D-22, F-6, the 2026-10-27 grouping rule ([2.0 / next] A held optimistic lane holds per render effect: latest(x) in its own JSX hole shows the guess beside held content (D-25) #4284: JSX grouping must never change the result).