Fix/pet card layout and smoke flake

[Cat5Dog2]

Two independent fixes, one per commit, touching disjoint files so either can be
reverted on its own.

1. fix: give the breed value the card width in the pet list (42aaf66)

On desktop the pet card laid its detail grid out in two columns, leaving each
value about 150px — not enough for a 12-character breed such as
スコティッシュフォールド, which wrapped mid-name. Breed is free text, so any
fixed column width fails again as soon as a longer name is entered.

Each item is now a single label | value row. Measured on the real page at
1280px (920px shell, three columns, 283px cards):

	before	after
breed value column	129px	173px
card height (desktop)	474px	490px
card height (390px)	643px	523px

Row spacing drops from 10px to 6px so the doubled number of separators does not
inflate the card; values also line up across rows and across cards.

This commit also replaces system-ui with an explicit Japanese font stack.
system-ui resolves to Yu Gothic UI on Windows, whose kana advance is about
0.72em, so identical text wrapped differently there than on macOS/Linux, where
it is 1em — スコティッシュフォールド measured 139px against 204px. Naming the
fonts narrows that gap (Windows now measures 192px). It is an approximation,
not a guarantee: where none of the named fonts exist it still falls back to
sans-serif.

Accepted trade-off

At three columns the value column is 173px, i.e. room for about ten full-width
characters. Longer breeds still wrap — now consistently on every OS rather than
only on some. Fitting twelve characters would need cards of 312px or more, which
means either a wider shell or dropping to two columns; three-column density was
chosen instead. The reasoning is recorded in the CSS.

Test

PetList_GivesBreedValueMostOfCardWidth_AcrossViewports fixes the width split —
the value keeps room for at least ten full-width characters, on the same line as
its label — rather than a character count, so it does not depend on which fonts
the CI runner has installed. It was confirmed to fail both on the previous
layout and on a two-column card grid.

2. fix: stop the local-smoke readiness test flaking under load (133b7c3)

readiness waits through 500 until the app returns 200 failed intermittently
with Timed out waiting for http://smoke.invalid (last status: 500).

wait_until_available polls against a wall-clock deadline and slept 0.5s
between attempts. The test queues five 500s before the 200, so 2.5s of the 5s
budget went to sleeping, and spawning the stubbed curl six times cost another
~1.5s on Windows — about 1s of headroom, which a busy machine ate.

The suite already states at the top that nothing in it waits on real time (it
zeroes SMOKE_RETRY_DELAY_SECONDS), but readiness had no such knob. This adds
SMOKE_READY_POLL_SECONDS, defaulting to 0.5 so CD behaviour is unchanged, and
zeroes it in the tests: the readiness phase drops from ~4.0s to ~1.46s. The
test's --timeout also goes from 5 to 20, because the remaining cost is process
startup and cannot be zeroed; nothing here asserts a readiness timeout and the
loop returns as soon as the queue yields 200, so the larger budget costs no run
time. A static guard for the new default is added alongside the existing one for
the retry delay.

Test plan

• scripts/build.ps1 — 0 warnings, 0 errors
• scripts/test.ps1 — 240 unit tests, 8 shell tests
• Playwright E2E — 9 tests
• scripts/format.ps1 — clean
• Layout swept over eight pages at 390px and 1280px: no horizontal scroll and
  nothing spilling out of the shell
• Smoke suite run five times standalone and three times while a dotnet test
  run loaded the machine: 8/8 every time

Review notes

Per AGENTS.md this touches none of the areas that require human sign-off
(authorisation, existence hiding, returnUrl, image handling, deletion). The
changes are CSS, a shell script default, and tests.