Benchmark and optimize sparse matrix assembly kernels

[ddement]

Task: Benchmark our new sparse matrix assembly kernels on CPU and GPU for a few different problem sizes. Perform optimization as necessary.

Why: The current implementations of these assembly kernels are focused on correctness, but may or may not be leaving performance on the table. This story is to figure out if they can be improved and improve them.

Done means: Benchmarks have been performed on CPU and GPU for both a long running 3 blade problem with constraints and a 3000 blade problem with constraints. If the new assembly stages are a significant contributor to the runtime, optimizations are applied to fix that.