[llmassisted]

What this PR does

Restores the Android Oura Ring 4 integration that was present in the previously shipped fork APK but absent from the upstream-based branch merged in #1. The missing commits removed the native-versus-NOOP HRV comparison, rotating-address connection recovery, paced Bluetooth writes, battery-response handling, and sleep-window fallbacks.

This merges oura-ring-enhancements into current main and resolves the overlaps with the newer upstream implementation:

• Restore the ring's battery reply decoding and paced/retried GATT command queue; retain the newer authentication watchdog, timestamp-anchor handling, and serial redaction.
• Rediscover the paired ring's current privacy address on reconnect, preserve stored credentials, route Settings reconnect to Oura, and request Bluetooth permissions from the ring's Live controls.
• Restore native Oura RMSSD beside NOOP-calculated HRV, including the separately labelled resting-window comparison. Native values do not replace NOOP's canonical HRV/Charge inputs.
• Restore Ring 4 native-resting and manually marked sleep windows, while retaining upstream HR-only detection and WHOOP source isolation. Accept an active Oura hypnogram even when its owner equals the imported namespace.
• Follow the active serial identity in Health data flows, restore ring-specific history syncing and foreground-only live HR demand, and change the analysis watermark so existing stored data is re-scored.
• Increment Android versionCode to 538. Staging package and signing identity remain unchanged.

Added: Test Centre SpO2 feature-mode write (Android)

Ports the Swift Test Centre feature-mode experiment (ryanbr#2105, OURA_PROTOCOL.md §7.5) to Android, scoped to SpO2: Enable writes mode automatic (2f 03 22 04 01), Disable writes mode off. Each is behind a confirmation, is only sent while the ring is idle-streaming, is followed by a status re-read, and is never sent automatically. Beyond the Swift version:

• The ring's 2f 03 23 <id> <status> reply for any feature other than daytime HR is routed as a verdict (accepted / NOT_SUPPORTED / NOT_AVAILABLE) and logged, instead of being handed to the driver as a no-op live-HR ack. Live-HR triplet behaviour is unchanged (pinned by tests).
• The first raw 0x8b spo2_r_pi payloads per session are logged as fixtures, since a Ring 4 is expected to emit that tag once SpO2 runs and NOOP has never received it.

Platform scope: this restores previously shipped Android-only fork behavior. Swift is unchanged; the Oura-specific window/UI additions are not presented as cross-platform parity work. No schema migration or deletion of stored health data is introduced.

Added: live HR lifecycle and HRV comparison fixes (04cd3af)

• Live HR demand now follows the main Activity's own lifecycle, so a cold launch (observer attached after resume) requests ring HR immediately.
• The NOOP-side HRV fallback reads R-R per resting window, so older nights can no longer consume the read cap for recent ones.

Added: Oura activity estimate from the ring's MET stream (b44d0a9)

• Each anchored 0x50 record is persisted as an OURA_MET event row (met_x10 exact tenths, sec_per_sample, state); previously it was only logged, and the ring trims its copy after each drain. Never scored, never a step count.
• The pre-existing on-device activity sidecar (oura-activity-*.jsonl, written since Oura: capture Tier-B activity MET stream to a JSONL research corpus (Swift + Android) ryanbr/noop#676) is imported once into the same rows; the (deviceId, ts, kind) PK dedups it against live writes.
• Kotlin port of the Swift OuraActivityEstimator; byte-identity pinned by an oracle test against the Swift source compiled standalone over 300 inputs.
• Health screen card (only when OURA_MET rows exist): today's active minutes (≥ 3 MET), MET-minutes and estimated above-resting kcal, plus the last 7 days, labelled as an estimate. Within-record sample order is unpinned and documented (±~13 min).

Type of change

• Bug fix

How it was tested

• Both Android debug and release Kotlin compilation passed.
• Full local Android suite: 6,445 tests ran. The four new Oura restoration scenarios passed. One source-layout assertion was subsequently replaced with a production-selector behavior test; two unchanged RecoveryDriversTest exact floating-point assertions fail on this Mac/JBR 21 runtime (-0.4999999999999929 versus -0.5).
• Final focused Oura tests and signed staging packaging are running. GitHub's Linux/JDK 17 full suite will verify the final commit.
• Source-hygiene gate and protocol documentation checks passed.
• New OuraSleepRestorationTest covers native-resting sleep and vitals without gravity, active-owner saved hypnograms, manual-window precedence/separate naps, and rejection of short resting windows as automatic sleep.
• Restored tests cover native HRV aggregation, active serial-id flows, GATT queue retries/pacing, stored-key selection, and ring generation/identity.

Hardware validation (Oura Ring 4, ORE_06, builds 535/536):

• Commands are accepted by Android and the ring; battery reads 100%; a full history drain completed (bytes_left 0) and 14 of 20 nights now score (was 4).
• The resting-HR automatic restore is accepted at connect (feature-mode reply feature=0x8 status=0 (accepted)), and daytime live HR is paused when no live screen is visible (dhr_disable).
• SpO2 write: on a ring reporting SpO2 mode=0 (off) (server-gated, never cloud-enabled), 2f 03 22 04 01 returned status=0 (accepted) and the re-read showed mode=1 (automatic); the mode persisted across a force-stop and fresh connection. Whether SpO2 records then arrive (and as 0x6F or 0x8b) awaits an overnight capture.

Earlier: the supplied diagnostic log was inspected. It records failed writes, repeated saved-address connection failures, and stored HR/R-R streams rejected as NO-NIGHT with no motion. No physical-ring test has been performed for this branch; pairing, battery refresh, overnight collection, and the restored screens still need on-device confirmation. No raw diagnostic log or personal health data is committed.

Related issues

Refs #1