Revise AIFS ENS notebook examples for diversity and visual impact

[aldenks]

Follow-up to #34. Replaces the example mix to improve visual impact and avoid overlap with other notebooks.

Changes

• First time series → Ulaanbaatar 2 m temperature. Replaces precipitation at Hilo. Deep-winter forecast showing a sharp warm-up, a plunge back to -25 °C, and a wide range of late-forecast outcomes (-32 °C to -9 °C by day 10). A new city not used in other notebooks.
• Removed the redundant "control member highlighted" cleaner-view plot.
• Quantile bands → Buenos Aires. Moves from Berlin and brings a Southern Hemisphere example with Pampas cold-front passages widening and narrowing the bands.
• Probability map → Hurricane Erin (Aug 2025). Storm Nils was already in the DWD notebook, and the 10 m wind threshold made the SW France coastline look like a data artifact. Erin is fully offshore and shows a clean ensemble cone of tropical-storm-force wind probability at a 72 h lead.
• Animation → ensemble-mean downward shortwave radiation. Picked specifically because it starts out sharp (crisp cloud-shadow patterns) and visibly blurs over the 10-day forecast as members diverge on cloud position, while the day/night terminator stays sharp.
• Closing section renamed to "A few ideas to explore." Dropped the incorrect "same 0.25° grid as IFS ENS" claim.

File size: 5.3 MB → 1.6 MB.

Test plan

• uv run .internal/run_notebooks.py ecmwf-aifs-ens-forecast.ipynb runs clean
• Visual review of plots and narrative

🤖 Generated with Claude Code