Android: mergeSleep richness exception — stage-less import yields to computed stages (parity with #241)

[ryanbr]

Re-review follow-up to the merged #241 + #246. Closes the last parity gap from that pair.

The gap

#241 added a richness exception to Swift WhoopStore.SleepMerge.merge — a stage-less import must not blank a computed day that has stage data. But the direct Kotlin twin, WhoopRepository.mergeSleep (whose own comment says "Mirrors WhoopStore.SleepMerge (SleepMergeTests)"), still had the plain imported-wins logic:

val importedDays = imported.mapTo(HashSet()) { endDay(it) }
out.addAll(imported)
for (s in computed) if (endDay(s) !in importedDays) out.add(s)

So after the merge, the two shared-logic twins diverged: on iOS a stage-less WHOOP/Apple re-import yields to computed stages; on Android it still clobbered them.

Why #240 doesn't cover it

#240 (still open) fixes the Android side a different way — a Health-Connect-specific coveredDays write-gate plus purgeHcShadowed* heal that deletes sparse HC rows. It never touches mergeSleep, and it only guards Health Connect imports — a sparse Apple/CSV import would still clobber computed stages at the merge. This PR is the mergeSleep-level twin that matches Swift for all import sources.

The fix

Port the exception into mergeSleep (group by end-day; if the import for a day has no stages and the computed day does, the computed sessions win — otherwise imported wins, unchanged). Adds hasStages() with byte-identical semantics to Swift (null / "" / "[]" / blank = no stages).

Tests

Mirror Swift SleepMergeTests' 5 richness cases in MergeSleepLocalDayTest (stage-less→computed, import-with-stages-wins, neither-side, []/blank stageless, keep-every-session-of-winning-day). compileFullDebugKotlin + MergeSleepLocalDayTest pass (8/0). Single-day imported-wins and #715/#304 behaviour unchanged.

---

Re-review update (2nd commit): the first commit fixed mergeSleep (CSV export + HealthConnectWriter) but the Sleep screen builds its night list with its own inline imported-wins merge (two sites) that had no richness — so the actual sleep display would still blank stages on Android while iOS doesn't. Extracted mergeSleepRichness and routed both SleepScreen sites through it (preserving their effectiveStartTs sort). All three sleep consumers now share one richness rule. Tests 9/0.