feat: ADR-080 P1+P2 remediation — refactor, perf, tests, safety

P1 fixes (this sprint):
- P1-6: Extract sensing-server modules (cli, types, csi, pose) from main.rs
- P1-7: DDA ray march for tomography — O(max(n)) replaces O(n^3) voxel scan
- P1-8: Batch neural inference — Tensor::stack/split for single GPU call
- P1-10: Eliminate 112KB/frame alloc — islice replaces deque→list copy

P2 fixes (this quarter):
- P2-11: Python unit tests for 8 modules (rate_limit, auth, error_handler,
  pose_service, stream_service, hardware_service, health_check, metrics)
- P2-13: MAT simulated data safety guard — blocking overlay + pulsing banner
- P2-14: Wire token blacklist into auth verification + logout endpoint
- P2-15: Frame budget benchmark — confirms pipeline well under 50ms budget

Addresses 8 of 10 remaining issues from QE analysis (ADR-080).

Co-Authored-By: claude-flow <ruv@ruv.net>

v1/src/api/middleware/auth.py

@@ -189,7 +189,11 @@ class AuthMiddleware(BaseHTTPMiddleware):
                 self.settings.secret_key,
                 algorithms=[self.settings.jwt_algorithm]
             )
-            
+
+            # Check token blacklist (logout invalidation)
+            if token_blacklist.is_blacklisted(token):
+                raise ValueError("Token has been revoked")
+
             # Extract user information
             user_id = payload.get("sub")
             if not user_id: