[pipiche38]

Stacked on #2426. This branch contains #2426's commit (c7a678c32) plus one commit (ea58148ef). Review the second commit; once #2426 merges, this diff reduces to it alone.

Problem

This is the second half of #2425. The always-on hrv diag line pooled every sleep session of the day and divided their beat-time by first-to-last beat, so the hours between a night and a nap entered the denominator. On a real day whose main night the #1118 gate refused at 1.31 coverage, the line printed:

hrv diag day=2026-09-20 … coverage=0.48 … rrIntegrity=underCovered

The log called a night refused for an over-count "not an over-count", and it sent triage of that report the wrong way first. Two other readers use the same rows: hrvOverCounted (the HRV card's "unverified" flag) and the RSA respiration gate. Both sided with the log against the gate.

Change

• analyzeDay exposes the main-night group it already scores from, as DayResult.mainNightBlocks (the same mainGroup, not re-derived). It's trailing and defaulted on both platforms, and each has a single construction site.
• A new pure helper, AnalyticsEngine.hrvDiagnosticRows, selects that group's beats over the gate's own inclusive [start, end] window. On a one-block night that's exactly the set of beats SleepStager.sessionHrvOverCounted judged, so the over-count half of the verdict cannot disagree with the gate.
• A day with no main night keeps the old pooled, half-open set.
• Swift and Kotlin IntelligenceEngine read the rows through the helper. The line's format is unchanged; its fields now describe the main night.
• Known limit: a bridged multi-fragment night still spans its short bridge gap, which can only pull coverage down.

Verification

• HrvDiagnosticRowsTests (Swift) and its Kotlin twin HrvDiagnosticRowsTest, same fixtures and expected values:
  • DayResult carries the scoring main group.
  • On a refused night, the selected rows are the night's alone and classify as an over-count, agreeing with the gate. The old pooled selection on the same day reads plausible, which is pinned so the test fails if the fixture stops reproducing the disagreement.
  • With no main night, the half-open fallback is kept.
• swift test StrandAnalytics: 2056/0. Kotlin twin: 3/0.
• Full Android unit suite: the same 8 failures as untouched upstream/main (RecoveryDrivers, StandardHrSensorFormat, StepsDetailIntegration, TodayExplainability), none in touched code.
• App-target Swift changed (IntelligenceEngine.swift), so both app targets were built locally: macOS Strand and iOS NOOPiOS (generic Simulator) both build.
• doc_comment_lint: OK. No UI strings.

Refs #2425, #1118