[WIP] Callback-based synapses

[wsphillips]

This is a major overhaul of the package to support event-based synapses in addition to continuously integrated synapses. Up until now, there was some ad hoc "tricks" used to support "events", but this was not composable or robust.

When this PR finally lands, we should be able to:

• Detect spikes either discretely (at the end of each time step) or continuously--rolling back the solver to find the exact spike time of each neuron. Built-in spike detection will be based on positive-going threshold crossing, but I am designing the API to support alternative spike-event conditions (ie. roll your own methods)
• update/perturb the state of synaptic conductance models by summing (for linear changes) or independently (for models where the perturbation size is state-dependent)
• mix different types of synaptic model responses (spike detection must be totally discrete or continuous)
• callbacks will also handle hooks like threshold-based voltage resetting

[wsphillips]

test of callback based event-driven synapse (with just single exponential decay kinetics)

[wsphillips]

An updated version of demo/simplesynapse.jl shows how we can now flexibly change between continuous (root finding) events, discrete (end of time step) events, and continuously integrated synaptic current models. I will propagate the changes to the other demos and polish some minor lingering issues. It should be really interesting to see how the 100 neuron Pinsky & Rinzel network behaves upon switching over!