fix(studio): memoize only a launch that was actually attempted

[KnockOutEZ]

Closes wigolo-studio-run#185 (PX0 exit-4, K1/K3/K5 + coverage gap 2).

What was wrong

ensureStudioRunning's memo was written on the shared promise from a bare null, which cannot
distinguish a launch that ran and produced no handle from one that never started. A launcher
DECLINE — the normal outcome when substratePresent()'s 5s TTL disagrees with defaultLaunch's
uncached record read — was therefore remembered for 60s, so a substrate reinstalled a second later
was declined with zero spawn attempts: the lockout the memo's own docstring forbids.

Three siblings landed with it: memoMs: 0 disabled the write but not the read, so the documented
escape hatch was silently declined by another caller's window; normalizeLaunch read any object as
an outcome, so an async launcher's Promise (.started === undefined) became a silent instant
decline with the poll skipped; and DEFAULT_NO_HANDLE_MEMO_MS was unpinned — raising it to an hour
left all 103 arms green because every memo arm passed memoMs explicitly.

What changed

• The launch pass reports { handle, attempted }. Only attempted && !handle is memoized, which
  makes the comment above the gate true. A launcher that threw counts as not attempted for the same
  reason a decline does — nothing started, nothing waited for, re-attempt costs one throw.
• memoMs: 0 is honoured at the read as well as the write. It does not disable single-flight.
• normalizeLaunch throws on a thenable, and on an object with no boolean started.
• The shipped 60s window is pinned behaviourally (a caller walked across the boundary), not by an
  equality assert.

Decisions recorded in DECISIONS-AUTO as A-185-1 (read-honouring for memoMs: 0) and A-185-2 (a
throw is not memoized), each with its reversal condition.

Demo — red before / green after

Arm (a), against the tip before the fix:

 FAIL  |unit| tests/unit/studio/auto-launch.test.ts > a launch that produced no handle is
   remembered briefly > does not memoize a LAUNCHER decline — a substrate reinstalled a tick
   later is retried
AssertionError: expected "vi.fn()" to be called 1 times, but got 0 times
 ❯ tests/unit/studio/auto-launch.test.ts:595:22
     593|       const launch = vi.fn(() => { publishHandle(); });
     594|       const h = await ensureStudioRunning({ dataDir: dir, launch, laun…
     595|       expect(launch).toHaveBeenCalledTimes(1);

      Tests  5 failed | 29 passed (34)

After: Tests 34 passed (34).

Mutation probe for the coverage arm — DEFAULT_NO_HANDLE_MEMO_MS set to 3_600_000:

      Tests  1 failed | 33 passed (34)
     × pins the shipped default window — long enough to spare one fan-out, short enough not to lock out

Constant restored by reverse-edit; suite green again.

Verification

npm test              → Test Files 955 passed | 5 skipped (960)
                        Tests 12054 passed | 20 skipped | 7 todo (12081)   exit 0
npx tsc --noEmit      → exit 0
npm run gate:studio   → exit 0

CI deferred — quota.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 5a476c40-8a98-4dee-93ad-431e13608d5c

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}