This WIP PR introduces a new educational example: plot_recommender_with_ranking_metrics.py under examples/neighbors/.

The example shows two approaches to content-based recommendation system:

1. Cosine similarity on item-item matrix
2. NearestNeighbors with cosine distance

It also uses top_k_accuracy_score for evaluation.

We’re splitting the example into two parts to show both the manual cosine similarity approach (for conceptual clarity) and the NearestNeighbors method (for scalable implementation).

The thinking is that this split will allow users to understand the mechanics while also learning an almost production-friendly pattern for mid sized datasets, while deeply understanding the difference between the two also on the performance level.

(too ambitious maybe? thoughts?)

✅ Plan:

• Add example docstring and scaffold
• Add/Create toy dataset
• Implement cosine similarity recommendation
• Implement NearestNeighbors recommendation
• Evaluate both with top_k_accuracy_score
• Light visualization (if needed)

Feedback welcome on the:

• overall structure
• location (examples/neighbors)
• approach on weather it makes sense to split into two or should we just go with one method in this example

cc @adrinjalali @StefanieSenger @lorentzenchr

ps. apologies on the previous pr open (rebase issue)