feat: per-step supervision over full sequence

- model outputs (B, S, 2) instead of (B, 2) — GRU output at every timestep
- train/val loss computed over all S timesteps with reduction=none
- benchmark/evaluate.py takes pred[:, -1, :] for final-step evaluation
- added per-step loss logging (8 evenly spaced steps) to TensorBoard

Generated by Mistral Vibe.
Co-Authored-By: Mistral Vibe <vibe@mistral.ai>

benchmark/evaluate.py

@@ -114,7 +114,8 @@ def evaluate_scene(
         tilt = batch["tilt"].to(device)
         target = batch["v_body_target"].to(device)  # (B, S, 2) normalized
 
-        pred = model(events, tilt)                   # (B, 2) normalized
+        pred = model(events, tilt)                   # (B, S, 2)
+        pred = pred[:, -1, :]                         # (B, 2) — last timestep
         target_last = target[:, -1, :]               # (B, 2) normalized
 
         all_preds.append(pred.cpu().numpy())