Evaluate Concurrent NDT, Gyro Odometry, and EKF Processing with Replay #758

Establish a reproducible replay-based evaluation of the integrated Awkernel localization pipeline, with NDT, gyro odometry, and EKF processing active concurrently. Evaluate scheduling interference, temporal correctness, and localization behavior under representative workloads.

Scope:

• Replay timestamped LiDAR, IMU, and vehicle velocity data while preserving their relative timing and input rates
• Run NDT, gyro odometry, and periodic EKF processing concurrently using documented task and CPU assignments
• Include the EKF-to-NDT initial-pose feedback path with causal data availability
• Compare a low-interference baseline with concurrent execution under increasing NDT workloads
• Trace input release, task readiness, execution, measurement consumption, and output publication
• Record queue occupancy, dropped inputs, convergence outcomes, and localization results

Acceptance criteria:

• Replay datasets, clock mapping, playback rate, task periods, deadlines, priorities, and CPU assignments are documented.
• Input release is independent of processing completion, so slow processing does not silently reduce the offered workload.
• Input and task traces confirm that the intended concurrent workload is exercised.
• Buffering and overflow policies are explicit, and delayed or dropped inputs remain visible in the results.
• Scheduling delay, execution time, deadline misses, output latency, and consumed-input age are measured with defined timing boundaries.
• Localization error and NDT convergence are evaluated alongside timing metrics, using ground truth where available.
• Repeated runs report variability and retain sufficient configuration and trace data for reproduction.
• Any differences from a reference run are analyzed with respect to both numerical behavior and measurement-consumption timing.

Out of scope:

Live sensor integration, closed-loop vehicle control, exhaustive scheduler comparisons, and systematic fault injection reserved for M3.