Add light classification + prefix sum before light culling job

[jeremyong-az]

Is your feature request related to a problem? Please describe.
Currently, lights are culled in a single uber-shader per 16x16 tile for all light types. This creates a worst-case occupancy characteristic for the light culling jobs.

Describe the solution you'd like
The typical approach is to classify each tile based on available lights and write out a mapping from tile classification to a set of tile coordinates, performing a prefix sum to determine a dispatch count per classification as well. This in turn drives a dispatch indirect per culling shader (one per classification).

Additional context
This is another optimization to consider along with #4330