Paretocurve

[Wenyueh]

title "Add Pareto frontier plots for model selection results"

Summary

• Add plot_pareto() method to SelectionResults that generates three pairwise 2D Pareto frontier plots: Accuracy vs Latency, Accuracy vs Price, Latency vs Price
• Axes oriented so "better" is always top-right, producing concave frontiers
• Points labeled with numbered indices (1), (2), etc. with a separate legend mapping numbers to combo names outside the plots
• Pareto-optimal points highlighted in blue with frontier line; dominated points in gray; best combo marked with a gold star
• matplotlib added as optional dependency under [plot] group, imported lazily
• results.plot_pareto() added to all 7 example scripts
• For bandit-type of problems, only graph the model combination that survive to the last stage

Test plan

• All 63 existing tests pass (uv run pytest -x)
• Run any example with --selector arm_elimination and verify the Pareto plot appears
• Verify results.plot_pareto("output.png") saves to file correctly

resolves issue #37