Refactor pixel snapping into a single solution?

[phargogh]

We have 3 functions that snap pixel coordinates to a raster's grid in similar but slightly different ways:

• warp_raster and create_raster_from_bounding_box both take the same approach
• align_bbox (used by zonal_statistics, added in zonal_statistics: expand bbox intersection to align with grid #333 )

The approaches are similar but slightly different. Now that we have 3 use cases, it might be worth looking at whether it makes sense to refactor in a way that covers these 3 cases and reduces duplication.