Fix critical logic errors in distributed path tracer

[Copilot]

Code review identified several serious bugs preventing the distributed path tracer from functioning correctly: incorrect ray routing, off-by-one resolution errors, and NDC calculation using tile dimensions instead of global image dimensions.

Ray Routing Fixes

• ACCUMULATE stage rays dropped: map_ray_stage_to_queue doesn't handle ACCUMULATE—changed to m_batch_sender.enqueue_ray(ray, MASTER_ID)
• Direct lighting results misrouted: Added owner check before routing (matching existing object intersection logic)

// Before: always local
map_ray_stage_to_queue(best_ray);

// After: route to owner
if (best_ray.worker_id == m_worker_info.worker_id)
    map_ray_stage_to_queue(best_ray);
else {
    best_ray.type = models::ray_type::OWN;
    m_batch_sender.enqueue_ray(best_ray, best_ray.worker_id);
}

Resolution/Bounds Fixes

• Master resolution off-by-one: (max - min) → (max - min) + 1
• Master bounds in preprocessor: max_x: X → max_x: X - 1 (inclusive bounds)
• NDC using tile resolution: Workers calculated NDC from local tile size, causing all tiles to render same view. Added image_width/image_height fields to worker_info

Concurrency Fix

• Batch sender data loss on shutdown: Removed early break in flush_loop() that skipped final flush when m_terminate set

Initialization

• Added proper default values for cloud_ray fields (object_intersect_distance, alpha, type)
• Fixed uvec2 declarations incorrectly using fvec2 constructor

Original prompt

Review my code.

Compile all SERIOUS logic errors.

Focus on any issues pertaining to concurrency (path-tracer-core)
Focus on incorrect aws resources setup (path-tracer-preprocessor & cdk)
Focus on incorrect usage of aws resources such as sns & sqs (path-tracer-core & path-tracer-preprocessor)

Essentially, verify that the application will generate a scene and work

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.