fix(kida): keep a child write in the caller's context

[dangreen]

childOper wrapped its whole write path in untracked — the reads of the parent and the key, and the write back into the parent along with them. Only the reads need it.

The write does not create dependencies, so untracking it looks free. It is not: activeSub is also how the lifecycle queue decides whether the system is quiescent.

function settle(): void {
  if (
    !draining
    && pending.length
    && (
      activeSub === undefined
      || (activeSub.modes & (ScopeMode | LazyMode)) === ScopeMode
    )
    && !batchDepth
    && !flushDepth
  ) {

A write from a subscriber's own body triggers a flush when nothing else holds one open, and flush ends by calling settle. With the write untracked, activeSub is undefined at that moment, so settle concludes that nobody is running and drains the queue on the spot — and the mount listeners the flush woke fire inside the subscriber that wrote, halfway through its body. With the write left in the caller's context, settle sees a running subscriber and defers them to the real boundary.

The gate

const stopWriter = effect(() => {
  log.push('start')

  if ($a() === 1) {
    $a(2)
  }

  log.push('end')
})

where a second effect reaches a mountable signal only once that write has landed:

	log
before	start, mounted, end
after	start, end, mounted

Two gravestones go into agera's signal spec next to it, pinning what a plain signal already guarantees and what the child now stops obscuring: an effect that writes its own dependency is not woken by its own write, whether it writes directly or through untracked.

Size

nanoviews gains 4 B gzip / 3 B brotli on the average-usage bundle — two let slots the write needs outside the untracked block. That is enough to push Average usage (Brotli) past its pin (3998 → 4001 B), so the limit moves 4 kB → 4.05 kB. Every other pin in the chain stays where it was.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 84.83%. Comparing base (23b3370) to head (5b8eee9).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #198      +/-   ##
==========================================
+ Coverage   84.73%   84.83%   +0.09%     
==========================================
  Files         140      140              
  Lines        3125     3125              
  Branches      586      586              
==========================================
+ Hits         2648     2651       +3     
+ Misses        339      338       -1     
+ Partials      138      136       -2     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.