Synthetic_Philly Optimization and Rework

[santosl529]

#Summary

• Add a reworked synthetic Philadelphia pipeline with cached spatial/raster city data, batched joblib generation, and Parquet batch outputs rather than DataFrames to reduce overhead
• Optimize destination diary generation with NumPy-backed EPR bookkeeping, candidate-only gravity lookup, and restored baked gravity callables from the city cache.
• Reduce dense trajectory overhead by avoiding unused Shapely route-boundary construction, reusing path-relative position while an agent follows the same route, and validating street coordinates with a precomputed set.
• Update the active Philadelphia bounding box to include more of West Philly.
• Increase number of agents from 200 to 1000.

Performance

Before batched joblib generation, agent, trajectory, and diary generation took 1709.27s for a 200-agent, dt=0.5 run. After parallelization, generation time improved to about 291s, then to about 104s after the diary/gravity/cache optimizations.

A single-agent dense trajectory profile improved from about 7.43s to 5.11s after the route-overhead cleanup.

Notes

After the bounding-box change, to fix the cached city, rerun once with:

REGENERATE_SPATIAL_DATA = True
REBUILD_CITY_CACHE = True

And then revert back to False after first run.

Closes #298