ENH: location generation improvements

[hongyeehh]

Our location_generation relies on the DBSCAN algorithm and enables the generation of user- and dataset-level locations. Locations are generated from core points from the DBSCAN results. When the input epsilon is small and locations only consist of 1 or 2 staypoints, we create a buffer with epsilon distance around the original point or line geometry to create valid polygon geometry for the column extent. This creates inconsistencies in the location results. For example, we create the following locations using epsilon=20 and num_samples=2:

• a. locations created from multiple sps are directly connected
• b. locations created from 2 sps are buffered

I suggest also buffering case a geometry to ensure consistency.

In addition, we occasionally observe huge locations created in dense sp regions, which overlap with many other locations (case c in the figure). We would need to check how these locations are generated and propose solutions to split/avoid them