Add Porto taxi trajectory rasterization example with custom log-sum merge

[brendancol]

Closes #1021

Summary

• Adds examples/user_guide/33_Porto_Taxi_Trajectories.ipynb with line rasterization on real GPS taxi trajectories
• Downloads 200K trips from the ECML/PKDD 2015 Porto taxi dataset (UCI ML Repository, ~509 MB). Each trip is a full GPS trace with readings every 15 seconds, not just pickup/dropoff points
• Parses polyline JSON into ~197K LineStrings, computes trip duration from GPS point count
• Rasterizes onto an 800x600 grid three ways: trip count, total duration (sum), and a custom @ngjit log-duration merge
• The log merge compresses dynamic range so side streets show up alongside major highways

Test plan

• Run the notebook end-to-end (requires ~509 MB download from UCI)
• Verify all three raster panels render on a dark background
• Check that ~59% of pixels have non-zero values (good trajectory coverage across Porto)