[ryanbr]

The problem

The Today heart-rate card's empty state said:

Calibrating, no heart rate banked yet today. Your curve fills in as the strap offloads.

The branch that renders it checks one thing: whether the active-strap-plus-imports union came back with fewer than two 5-minute buckets for the window. It does not know whether a strap is calibrating. It does not know whether a strap has offloaded. It does not know whether rows are banked under a source the union cannot reach. It asserted all three.

That matters because the card is the first thing a reader consults when a curve is missing, and naming an unchecked cause sends them to the strap when the answer may be somewhere else entirely.

The same line also fired on a day holding exactly one bucket, where "no heart rate" is simply false.

The change

The state splits in two and describes the stored rows:

day's buckets	line
0	No heart rate stored for today yet. The curve draws once the strap offloads, or once an import brings one in.
1	Only one five-minute block of heart rate is stored for today. The curve needs two of them to draw a line.

The past-day and narrow-window lines are unchanged.

The split reads the day's bucket count, never the windowed subset, because both new lines speak about today. A rolling 1h window that happens to exclude the one stored block must not make the card claim the day holds nothing, which would be the same over-claim in a new place. That rule has its own test.

Android's selection is extracted into a pure hrEmptyMessageRes(isToday, windowIsWholeDay, dayBucketCount) so it can be pinned without a Compose host. Apple's lives in hrEmptyTitle / hrEmptyDetail, hoisted out of the view body rather than nested as ternaries, since TodayView already sits near the type-check budget.

Localisation

today_hr_calibrating is replaced by today_hr_none_today and today_hr_one_block across the default resources and all eight Android locale files, so nothing is left behind to trip ExtraTranslation. Four new keys land in Strand/Resources/Localizable.xcstrings with all nine non-English locales filled, inserted in place rather than through a rewrite, so the catalog diff is 232 added lines and nothing touched.

Verification

• :app:compileFullDebugKotlin and :app:compileFullDebugUnitTestKotlin clean
• HrEmptyMessageTest, six cases: the past-day line regardless of count, the 0 and 1 cases, the window case, and the two that pin the day-count rule against a narrow window
• Source tripwires in the same test for the Apple twin and for the retired resource key, since neither has JVM reach of its own
• Tools/doc_comment_lint.py clean
• Tools/i18n_audit.py --ci origin/main clean, with no new focus-locale gaps on either platform

What this does not do

It does not make the card able to tell a strap that has never offloaded apart from rows banked where the union cannot see them. That needs a diagnostic the card does not have, and inventing one here would repeat the mistake. What it does is stop the card from guessing.

Refs #863