Rebuild TK5 support on the Yucheng YCBT protocol and add SmartHealth-app Colmi rings

[saksham2001]

Summary

The TK5's manufacturer supplied the SmartHealth Android app, whose Yucheng YCBT SDK is the
authoritative protocol. Comparing it byte-by-byte against our driver showed TK5 history sync was
fundamentally broken. This rebuilds the driver on the real protocol — and because Colmi rings that ship
with SmartHealth (rather than QRing) speak the same protocol, adds them as a second family off the
shared stack.

What was broken on the TK5:

• History was never actually requested. Our opcodes (02 24/26/28) are GetCardInfo /
  GetSleepStatus / GetMeasurementFunction. The real flow is 05 <key> query → header → data →
  terminal 05 80 → a mandatory ACK we never sent, so the ring never released the next data type.
• The "enable monitoring" burst was hitting the Health-DELETE opcodes (05 40…4E), so the ring was
  never told to record the metrics we then tried to read.
• Records were sliced per-frame, not from the reassembled block — silently dropping any record that
  straddled a GATT boundary. There was no fragment reassembly at all.
• Whole record types were never synced (SpO₂, BP, HRV/stress/fatigue, steps, temperature, blood sugar);
  setTime's weekday byte was hard-coded (the ring thought every day was Monday); the "user profile"
  was a frozen capture blob rather than the actual user.

New: a protocol-driven history state machine, GATT fragment reassembly, correct settings/monitor
commands, all 9 record types, device-push ACKs, and periodic re-sync. Capabilities are now decided by
the ring: the handshake reads its capability bitmap (02 01) and we only offer sensors it actually
claims.

Colmi + SmartHealth: a ring's Bluetooth name doesn't say which app it shipped with, so the pairing
screen asks; the choice picks the driver, and a wrong pick fails in 20s with a one-tap "try the other
app" retry.

The AE00 handshake is deliberately not implemented — it's JieLi RCSP and gates only OTA/watch
faces. Confirmed on hardware: a JieLi ring (chipScheme 4) answers every health command in plaintext.

Related issues

Part of #50 (TK5 A840 — this PR is the TK5 driver rebuild; that unit's TK5 A840 name matches the
TK5 <4 hex> pattern, but it hasn't been tested on that ring).

Type of change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• 📟 New / improved wearable support (BLE driver layer)
• 🤖 Coach / LLM change (tools, prompts, orchestration)
• 🎨 UI / DesignSystem change
• 🧹 Refactor / chore (no behavior change)
• 📝 Docs only
• ⚠️ Breaking change (existing data, settings, or APIs change)

_{UI: the pairing screen's app-type picker + the wrong-app failure/retry path. Not breaking: every
persisted enum is append-only and jring / QRing-Colmi source files are byte-identical, so existing
users' rings and data are unaffected.}

How was this tested?

• Added / updated unit tests (PulseLoopTests) — 612 passing (461 before this PR)
• Ran the test suite locally (⌘U in Xcode)
• Tested on a physical device with a real ring — model: Colmi R09 (SmartHealth, advertises R99 54DC)
• Tested with demo data (-seedDemo YES, no hardware)
• N/A (docs / non-code change)

Confirmed on the R09: pairing and app-type selection, plaintext handshake with no AE00 auth on a
JieLi chip, time sync (correct weekday), firmware + battery parse, the 9-query history sync with its
ACKs, live/spot HR, spot SpO₂, blood pressure + calibration, and workout HR/SpO₂.

The hardware session also caught two real bugs, now fixed: HRV was offered on a ring whose bitmap
denies it (it refuses the command outright), and the SpO₂ spot window (40 s) was shorter than the
ring's own sweep (38–41 s).

Not yet exercised on hardware: sleep (the multi-session timeline is decoded from the SDK and
unit-tested, but no YCBT ring has been worn overnight), and the TK5 itself — it shares this driver but
hasn't been re-tested, so it stays 🧪 Limited.

Privacy & data

• This change does not send health data off-device without explicit user action.
• No secrets, API keys, or personal data are committed.

Checklist

• My code follows the project's style (SwiftLint passes).
• I ran the tests and they pass.
• I updated docs / README where relevant — new docs/YCBT-Protocol.md (wire spec), rewritten
  hardware/tk5.md, and a SmartHealth section on hardware/colmi.md.
• I read the Contributing guide.