Add configurable auxiliary panel to unrolled grid plots

[AvencastF]

Motivation

• Provide an extra per-point plotting panel for metrics like effective_steps or min_val_loss so grid plots can show training-related auxiliary data alongside SIC, cutflow and winner panels.
• Make the auxiliary panel configurable and consistent with the existing plot configuration system and layout behavior.

Description

• Added an aux_panel configuration to DEFAULT_GRID_CONFIG (default enabled, metric_col: effective_steps, ylabel: Effective steps) and a new height_ratios.aux entry for layout control in paper_plot.py.
• Introduced _collect_grid_metric_map in paper_plot.py to gather per-point auxiliary metric series for each model and wired aux_by_model into the plotting call when aux_panel.enabled is true.
• Extended plot_unrolled_grid_with_winner_and_ratios in plot_styles/grid_unrolled.py to accept aux_by_model, reserve a subplot for the auxiliary panel in the shared layout, and draw per-model line plots for the chosen metric; values that are non-finite or <= 0 are ignored to match the requested behavior.
• Kept styling and behavior consistent with other panels: configurable marker/line options via aux_panel.line, optional log-scale via aux_panel.y_log, block separators applied, and axis style application includes the aux axis.

Testing

• No automated tests were executed as part of this change.

---

Codex Task