feat: complete ruv-neural implementation — physics models, security, witness verification
Replace all stubs/mocks with production physics-based signal models:
- NV Diamond: ODMR Lorentzian dip, 1/f pink noise (Voss-McCartney), brain oscillations
- OPM: SERF-mode, 50/60Hz powerline harmonics, full cross-talk compensation
via Gaussian elimination with partial pivoting
- EEG: 5 frequency bands, eye blink artifacts (Fp1/Fp2), muscle artifacts,
impedance-based thermal noise floor
- ESP32 ADC: ring-buffer reader with calibration signal generator, i16 clamp
Security hardening (SEC-001 through SEC-005):
- RVF bounded allocation (16MB metadata, 256MB payload)
- sample_rate validation (>0, finite)
- Signal NaN/Inf rejection
- ADC resolution_bits overflow clamp
- HNSW HashSet visited tracking + bounds checks
Performance optimizations (PERF-001 through PERF-005):
- 67x fewer FFTs via pre-computed analytic signals
- VecDeque O(1) eviction in memory store
- Thread-local FFT planner caching
- BrainGraph::validate() for edge/weight integrity
- Eigenvalue convergence early termination
Ed25519 witness verification system:
- 41 capability attestations across all 12 crates
- SHA-256 digest + Ed25519 signature
- CLI commands: `witness --output` and `witness --verify`
README: ethics warning, hardware parts list (AliExpress), assembly instructions
Co-Authored-By: claude-flow <ruv@ruv.net>
614d242967