feat: add Population Dynamics Insights custom boundary aggregation guide

[henrikvalv3]

Summary

This PR introduces a new notebook demonstrating how to aggregate Population Dynamics Insights (PDI) embeddings into custom, arbitrary boundaries (e.g., retail trade areas, 5km isochrones).

To prevent the spatial distortion caused by naive area-weighted averaging, this guide uses a population-weighted average using the WorldPop 2025 dataset.

Key Features/Notes

• Uses BigQuery S2_COVERINGCELLIDS and Earth Engine's ST_REGIONSTATS to perform all spatial overlays and population counting.
• L2 normalization has been decoupled from the SQL pipeline. The BigQuery output retains the raw signal magnitude (volume/density). An optional Scikit-Learn Python step is included to L2 normalize the vectors for downstream algorithms that require unit vectors (e.g., Cosine Similarity, K-Means clustering).
• Includes a testing cell using NumPy to validate that the resulting embeddings maintain exactly 330 dimensions and (if normalized) a magnitude of ~1.0.
• Added licensing information and the recommended academic citation for the WorldPop dataset at the bottom of the notebook.