[bhelm]

What this PR does

Makes the Swift public SleepStagerV2.stageSession establish stable timestamp order before binary-search clipping, matching the Kotlin twin. Already-monotonic streams keep the allocation-free fast path; only genuinely shuffled inputs allocate a stable reorder. The uncached recipe consumes the already-ordered clipped streams directly, preserving same-second emission order.

Permanent mirrored native tests cover deterministic shuffled streams, freeze the existing sorted-input result, and use cache-isolated translated windows to prove two order-sensitive same-timestamp gravity outcomes remain distinct while timestamp-group shuffling preserves each outcome. Padding constants and the parity rollout harness are unchanged.

Type of change

• Bug fix
• New feature
• Refactor / cleanup
• Documentation
• CI / tooling

How it was tested

• RED on unchanged 6df88d0d: the permanent Swift shuffled-input regression failed with sorted all-light versus shuffled light+REM.
• RED mutation check: a legal unstable same-timestamp tie reorder made both retained-signal oracle runs change from wake to light.
• Final focused Swift public outcome regression: passed.
• Final mirrored Kotlin public outcome regression: passed with JDK 17, one worker, Kotlin compilation in-process, and parallelism disabled.
• Production-identical full StrandAnalytics suite: 1,439 tests executed, 2 skipped, 0 failures.
• Production-identical full Android :app:testFullDebugUnitTest: passed.
• Clean current-main Kotlin SleepStagerV2Test: passed under the same resource limits.
• Clean current-main Swift test invocation is presently blocked before test execution by the unrelated Linux baseline import of the Apple-only Compression module in WhoopStore/RawOutbox.swift; the fork full Swift suite and focused final Swift regression are green.
• Range-diff from the reviewed fork commit to current main differs only in the already-landed Fix sleep cache gravity fingerprint collisions #1453 gravity-fingerprint context; behavior and tests are unchanged.
• Parity ratchet against the exact fork base: 59 differential, 2,605 exempt, 0 errors.
• Independent final delta review: no P0/P1/P2 findings.
• No manual or hardware testing required: this changes pure sleep analytics ordering only.

Checklist

• Swift package tests pass for any package I touched (swift test in Packages/<name>)
• Android unit tests pass if I touched android/ (./gradlew testFullDebugUnitTest)
• No new build warnings introduced
• UI changes use only StrandDesign tokens — no hardcoded colors, fonts, or spacing
• No hardcoded hex frame bytes; protocol facts live in the schema / decoders
• Follows the conventions in docs/CONTRIBUTING.md
• I did not commit generated output (Strand.xcodeproj/) or any secrets/keystores

Related issues

Tracked in bhelm#67.