Improve Efficiency of Hamiltonian usage in simulate_de

[zmorrell]

The majority of the memory allocations being performed in simulate_de seem to be coming from generating the hamiltonian in each iteration. Improving the efficiency (perhaps by precomputing the matrix components dependent on A(s) and B(s) and performing some sort of caching of the matrix used in the solve loop) could likely greatly improve performance, which would be beneficial for large system sizes.