Add val/train loss to evaluation and plots

- _evaluate() now computes per-batch loss when loss_fn is provided
- Val loss and train recall loss saved in val.csv and train_recall.csv
- Overview plot shows train vs val loss curves side by side
- Helps detect overfitting: val loss diverging from train loss

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

src/training/plot_metrics.py

@@ -114,12 +114,15 @@ def plot_combined(train_df: pd.DataFrame, val_df: pd.DataFrame, out_dir: Path) -
     fig, axes = plt.subplots(1, 3, figsize=(18, 5))
     fig.suptitle("Training Overview", fontsize=16, fontweight="bold")
 
-    # 1. Train loss.
+    # 1. Train + Val loss.
     ax = axes[0]
-    sns.lineplot(data=train_df, x="epoch", y="total", ax=ax, marker="o", linewidth=2, color="crimson")
-    ax.set_title("Train Loss")
+    sns.lineplot(data=train_df, x="epoch", y="total", ax=ax, marker="o", linewidth=2, color="crimson", label="train")
+    if "loss" in val_df.columns:
+        sns.lineplot(data=val_df, x="epoch", y="loss", ax=ax, marker="s", linewidth=2, color="royalblue", label="val")
+    ax.set_title("Loss (train vs val)")
     ax.set_xlabel("Epoch")
     ax.set_ylabel("InfoNCE Loss")
+    ax.legend()
 
     # 2. Val R@1 q→g.
     ax = axes[1]