Add / use cache-optimized indexed priority queue in NRM-based methods.

NRM and CoEvolve are often bottlenecked by the cost of updating next event times within their data structures. I suspect swapping out the indexed priority queue we use for another data structure could give much better performance. A reason this would be advantageous is that it seems like these are easier methods to use / implement with variable rates and delays compared to direct type methods. This ties into implementing the Sanft-Othmer method (https://arxiv.org/abs/1503.05832) which is already mentioned in another issue and is NRM via a table-based data structure.

Start by locating the indexed priority queue implementations used by NRM and CoEvolve, then read the Sanft-Othmer paper linked in the issue and the related issue on that method. Compare the current update costs with a cache-optimized table-based structure. Done should include an implemented replacement and evidence that NRM-based methods improve without breaking variable-rate or delay behavior.