Projection Optimization

[danieljvickers]

The current code projects particles one at a time, with a different kernel for each. This is horribly inefficient. We need more parallelism. Explore a refactor that chooses the parallelism based on the particle count and have them both call the same interior checking code.