Smoke and CI wall-clock budgets: the five rows #2323 left as "assess"

[cliffhall]

Split out of #2323, which enumerated every wall-clock budget in the repo's test
gates and fixed the Vitest, Testing Library and Playwright ones. Five rows in
its Group F/G were marked assess — they need a read that survey did not do,
and #2323 says so explicitly under "Follow-up work identified but not in scope".
The cheap, definite one (F6, hoisting the 19 scattered Playwright locator
budgets into scripts/lib/browser-timeouts.mjs) was folded in; these are the
rest.

Each needs the same two questions answered that #2323 answered for the others:
is this a ceiling on a poll or a fixed sleep, and is the number one
somebody chose for a machine running three or four concurrent worktree gates.

#	Site	Current	What to decide
F1	scripts/lib/announced-child.mjs:48-49	timeoutMs = 30_000, pollMs = 250	Bounds test-server readiness for every smoke that needs one, so it gates the widest surface of the five.
F4	scripts/lib/mcp-app-flow.mjs:177-179	goto 30_000, connect 45_000, ready 45_000	Predates browser-timeouts.mjs and duplicates its ui/roundTrip split by hand. Either adopt those constants or say why this flow differs.
F5	scripts/lib/deep-link-connect.mjs:60-61	goto 30_000, connect 45_000	Deliberately mirrors F4 — keep them shared, don't fork. Same call as F4.
F7	scripts/lib/pty.mjs:133	spawnSync(…, { timeout: 5000 })	Probes for script(1). A 5s bound on a process spawn under 8x oversubscription is thin, and its expiry means "no pty available" rather than "slow", so a false negative silently changes what the TUI smoke tests.
G1	.github/workflows/main.yml — no timeout-minutes on any job or step (5 jobs)	GitHub's 360 min default	Set an explicit per-job budget as a hung-job guard, not as a flake fix. Nothing in #2323's evidence implicates CI — GitHub runners are not the contended machine. sdk-watch.yml:119 already does this (timeout-minutes: 20) and is the precedent.

Not in scope

• Do not scale the ~95 fixed sleeps in test files. Test-gate timeouts are still library defaults — the unit, tui and launcher projects and all 788 Testing Library waits fail correct tests under concurrent-worktree load #2323 settled this: a
  setTimeout(r, N) with no condition always waits the full window, so raising
  it slows every passing run and still races on a loaded one. Replace one with a
  condition wait when it actually flakes (Flaky web test: ServerImportJsonModal debounce guard fails under a full parallel run #2250).
• Do not add retry. verify:test-timeouts enforces its absence, for the
  reason test-stability: eliminate timeout flakiness in the web test suite (zero-flake requirement) #1596 gives.

Acceptance

• Each of the five rows is either changed with a comment saying what the budget
  covers, or recorded on this issue as "keep, the default is correct" with the
  reasoning.
• F4 and F5 end up sharing whatever they end up with.
• If any Playwright budget moves, it moves in scripts/lib/browser-timeouts.mjs
  (or joins it), not at a call site.