feat(sensing-server): wire wifi-densepose-vitals 4-stage pipeline

[schwarztim]

Motivation

The sensing server's vital_signs module previously used an in-process FFT heuristic that produced jittery heart-rate readings (varying ±15 bpm minute-to-minute) and inconsistent confidence scores. The workspace already includes the wifi-densepose-vitals crate (4-stage IIR + autocorrelation pipeline), but it wasn't wired into the live REST path.

Design rationale

• Why a wrapper (CrateVitalsPipeline) instead of replacing the heuristic outright? Confidence-based fallback. The crate produces high-quality estimates only when the input signal has good SNR. Below confidence 0.05 the legacy heuristic outperforms the crate output, so we fall back. This preserves availability under poor signal conditions.
• Why threshold 0.05? Empirically calibrated against the test fixture set in wifi-densepose-vitals/tests/. Confidence below 0.05 corresponds to autocorrelation peaks indistinguishable from noise.
• Why call per-frame instead of batched? The legacy path is also per-frame; matching the call cadence simplifies the swap and avoids buffer-management changes.

Changes

• v2/crates/wifi-densepose-sensing-server/src/vital_signs.rs: add CrateVitalsPipeline struct wrapping wifi_densepose_vitals::VitalsPipeline; insert into extract_vitals() flow with confidence threshold; fallback to existing heuristic on low confidence.
• v2/crates/wifi-densepose-sensing-server/src/types.rs: replace VitalSignDetector with CrateVitalsPipeline in NodeState and AppStateInner (both struct fields and initializers).
• v2/crates/wifi-densepose-sensing-server/Cargo.toml: add wifi-densepose-vitals workspace dep.

Test results

• cargo test --workspace --no-default-features: 1632 passed, 0 failed
• python verify.py: pre-existing hash mismatch on this machine due to numpy/scipy version delta (hash generated on a different platform); our changes make no modifications to archive/v1/data/proof/ — confirmed by git diff origin/main..HEAD -- archive/v1/data/proof/

ADR / docs

• References ADR-045 (vitals jitter investigation); design rationale above is the implementation decision record for this wrapper.
• No new ADR needed; the wifi-densepose-vitals crate already has its own design notes.

[ruvnet]

Hi @schwarztim — current main has moved (mergeable=DIRTY/UNKNOWN). Key landings since this PR opened that you'll want to pull in:

• fix(vital_signs): use circular variance for wrapped phases #595 — vital_signs.rs now uses circular variance for wrapped atan2 phase values (relevant if you touch phase_var or the heartbeat FFT buffer)
• fix: bug triage for #559, #561, #588 #590 — fixes for the Fuzz Testing + Swarm Test (ADR-062) CI jobs (esp_stubs.h adds wifi_ps_type_t/WIFI_PS_NONE, scripts/qemu_swarm.py adds --force-partial); also corrected firmware README flash offset 0x10000 → 0x20000
• fix(security): host-header allowlist on sensing-server HTTP + WS (DNS rebinding) #580 — DNS-rebinding host-header allowlist on the sensing-server HTTP + WS routers (new host_validation.rs module wired into both routers)
• fix(sensing-server): handle WebSocket Lagged + add ping keepalive #484 — RecvError::Lagged now continues instead of disconnecting (WebSocket broadcast handler)
• v2: pin Rust 1.89 and fix sensing-server UI path when run from v2 #523 — v2/rust-toolchain.toml pins Rust 1.89; --ui-path default changed from ../../ui to ../ui with coalesce_ui_path() fallback
• chore(ci): add fix-markers for recent merges (#559, #561, #588, #593, #590-CI) #606 — scripts/fix-markers.json regression guard expanded — your PR shouldn't trip any markers, but the JSON's there to anchor the fixes above

The PRs were all evaluated thoughtful and well-described. Once you rebase against main, I'll do a focused review and merge if the test plan still passes locally. If you'd like me to rebase one of these on your behalf, say which (I can push back to your fork since maintainerCanModify is enabled).