perf: parse each live WHOOP4 frame once instead of 2-3× (thread ParsedFrame through router/collector)

[MrL0]

The problem

On the WHOOP 4.0 live path, a single reassembled frame is decoded by parseFrame 2–3 times — once per consumer — instead of once. During the type-43 realtime flood, each redundant decode re-runs loadSchema(), verifyFrame() (CRC32) and the FieldBuilder pass, on the same ingest path the recent decode work (rawHex fast-path, off-main decode) has been tightening.

Current flow (v8.3.4 / 25f1b41), for one frame in BLEManager.didUpdateValueFor:

1. router.handle(frame:) → FrameRouter.handle parses at FrameRouter.swift:24 to drive LiveState. Parse Import v8.2.2 snapshot (upstream catch-up + board wave; ports PRs #1035/#1036 + BLE/readiness fixes) #1.
2. While clockRef == nil, parseFrame(frame) at BLEManager.swift:2802 for clock correlation. Parse Fork CI: staging-release APK, versioned artifact names, R8-optimised release #2 (pre-clock only).
3. collector?.ingest(frame) (BLEManager.swift:2819) buffers raw bytes; Collector.flush() re-decodes the batch at Collector.swift:143 before extractStreams. Parse Revert release to unminified (R8 minify crashes at runtime); keep staging-release + versioned names #3.

Steady-state = 2 decodes/frame (router + flush); pre-clock = 3. (WHOOP 4.0 only — 5/MG doesn't ingest REALTIME_DATA into the Collector.)

The idea

Parse once at the BLE seam and thread the ParsedFrame through:

• In didUpdateValueFor, decode once: let parsed = parseFrame(frame, family: family).
• Add FrameRouter.handle(parsed:frame:) — keep a handle(frame:) shim that parses-then-forwards so existing callers/tests are unchanged; pass frame too, for the byte-level sub-decoders.
• Reuse that parsed for the clockRef == nil block instead of re-parsing at :2802 — byte-identical for WHOOP 4.0 (parseFrame(_:) == parseFrame(_:family:.whoop4)).
• Collector.ingest(frame:parsed:) buffering (frame, parsed) pairs (raw is still needed for the raw-capture outbox); flush() then calls extractStreams(buffer.map(\.parsed), …) instead of re-mapping at :143. extractStreams already takes [ParsedFrame] (Streams.swift:170), so the plumbing is there.

Output is byte-identical — same ParsedFrames, decoded once. Steady-state drops 2→1 decode/frame; pre-clock 3→1. No wire/command change.

Cross-platform: the Android side has the same shape — WhoopBleClient.handleFrame is a commented "direct port of FrameRouter.handle" and re-parses the batch in its flush path — so the same fix applies there too.

I'm raising this as an issue first (rather than opening a PR straightaway) because it's on the BLE path and the safety contract asks to discuss before touching Bluetooth. Happy to open the PR if you're on board — it'd just want a quick real-hardware sanity check (live HR/RR still update, fresh-connect still clock-correlates), since I can't strap-test it in my environment.

Surfaced via an AI-assisted perf pass; refs checked against 25f1b41.