Better support for solving decoders by simulating over time

[arvoelke]

Keeping a list of thoughts continuing off https://forum.nengo.ai/t/solving-for-decoders-by-simulating-the-neurons-over-time/385?u=arvoelke

• New approach needs to be able to have synapse=None for use in rate mode.
• Move override_rates_method into TemporalSolver.
• Approach taken for Reservoir Computing networks are too hacky (which is why they weren't ever officially released by nengolib).
• New approach needs to be able to include recurrent connections and simulate those too. Long-term: some sort of dependency graph of connections that depend on one another (and are then solved and simulated in that order).

[arvoelke]

Addressed by #137. I'll consider the longer-term goals of including RC/FORCE-style recurrent connections out of the current scope.