feat: add on-device Daily Coach briefings

[tigercraft4]

Closes #1736

What this PR does

Adds an opt-in, fully on-device Daily Coach briefing after overnight data has synced and been scored. It reports only the Charge, Rest, HRV, resting-heart-rate, and sleep values that are actually available, then adds a conservative training recommendation only when Charge exists.

The notification is disabled by default, uses no provider or network request, and is de-duplicated persistently to one notification per scored night.

Review feedback addressed

• iOS and Android now use the same metric names and rounded values: Charge, Rest, HRV, RHR, and rounded sleep hours.
• The displayed rounded Charge value is also the value used for the 34/67 recommendation boundaries, so the number and advice cannot disagree.
• A missing Charge no longer falls back to 50; other available metrics may still be shown, but the training line is omitted.
• Notification copy is routed through localized string resources on both platforms instead of new hardcoded English UI strings.
• Focused tests cover all metrics, both band boundaries, rounded-boundary behavior, missing Charge, and the all-metrics-missing case.

Type of change

• New feature

How it was tested

• python3 Tools/i18n_audit.py --ci origin/main — passed.
• xcodebuild -project Strand.xcodeproj -scheme NOOPiOS -destination "platform=iOS Simulator,name=iPhone 17 Pro" CODE_SIGNING_ALLOWED=NO build — passed locally.
• ./gradlew :app:compileFullDebugKotlin — passed locally.
• ./gradlew :app:testFullDebugUnitTest --tests com.noop.notif.ScheduledReportPolicyTest — 18 tests passed locally, 0 failures/errors.
• GitHub Actions on exact head 0eebd98b: 5/5 checks passed, including the NOOPiOS and macOS builds, Android build/tests, i18n Coverage, and Source Hygiene.

No real-strap test was run: this change consumes already-scored local values and does not alter BLE collection or protocol behavior.

Checklist

• Relevant iOS and Android builds/tests pass
• No new build warnings introduced by this change
• UI changes use existing design components/tokens
• No BLE protocol bytes changed
• No generated project output or secrets committed

[ryanbr]

The feature is sound and genuinely on-device — opt-in, no provider call, de-duplicated per night. The shared copy and the band input need work before it can land.

The two platforms brief differently. Same feature, different words and different metrics:

	Android	iOS
recovery label	Charge 72	Recovery 72
Rest score	included	absent
sleep 7 h 59 m	Sleep 7 h	Sleep 8 h

"Recovery" also contradicts iOS's own vocabulary — MetricCatalog.swift:132 names that metric Charge — so the notification disagrees with every other iOS surface. And the sleep line is a byte-level divergence in a displayed value: Android's it / 60 is integer division, silently dropping up to 59 minutes, while Swift rounds.

Swift shows one number and advises on another.

parts.append("Recovery \(Int(recovery.rounded()))")      // 66.6 → "Recovery 67"
switch recovery ?? 50 { case 67...: … case 34..<67: … }  // 66.6 → mid band

A 66.6 night renders "Recovery 67" beside the advice for the band whose ceiling is below 67. Android bands an already-rounded Int, so the same night gets the other recommendation.

A missing recovery is guessed as 50, against both files' own contracts. chargePct ?: 50 / recovery ?? 50. The Kotlin doc block two lines above says absent values are "never shown as 0 or a guess"; the Swift file header says "a missing metric is omitted instead of guessed". The consequence isn't cosmetic — a night with no Charge but some HRV still emits a training recommendation derived from a recovery that does not exist. Omitting the training line when there is no Charge would keep both promises.

The English is invisible to CI. No strings.xml, no xcstrings. The Android literals sit in a non-@Composable object, which Tools/i18n_audit.py does not scan — I ran it on this branch and it passes green while "Daily Coach", the three "Training: …" sentences and "HRV $it ms" / "RHR $it bpm" / "Sleep … h" ship untranslated. The file already had English literals so the debt is pre-existing, but this grows it from two short phrases to a title plus three full sentences.

Test coverage is one changed assertion. Three new bands, three new metrics, a guess path and a truncation, and the only test change is title.contains("recap") → title.contains("Daily Coach"). Worth pinning the 67/34 boundaries and that a missing Charge produces no training line.

One process note: the body says Closes #1736, which will auto-close that issue on merge. If the issue should outlive the PR, write the number bare.

Thanks @tigercraft4 — the privacy shape here is right, and the de-dup key per scored night is the correct design.

[tigercraft4]

@ryanbr this is now rebased onto current main (a8e88d71) and your review points are addressed: shared Charge/Rest copy and rounding, no fabricated Charge recommendation, localized notification copy, and focused boundary/missing-data tests. The exact head 0eebd98b is conflict-free and all 5 CI checks are green, including iOS, macOS, and Android. Ready for your merge review.