TODO
----
1) Maybe: first test with cylinders in Neuron, check access resistance and all.
	- Cylinder = fine
	- Tapered cylinder = almost fine (slight difference at the start, maybe different boundary condition?)
1b) Remy's stuff: does *not* work (maybe the gpu version?)
1c) Write it myself, use Remy's examples to check.
	- Works for 1 branch.
	- Impulse response works for sequence of branches.
	- Ball and stick works! (same as Neuron).
	- Hack for soma works.
	- Several branches: seems to work.
	(- Then add soma more cleanly.)
2) Check with active currents.
	- HH on cylinder, matches nicely.
	- HH on soma+axon+dendrite works.

* Threshold, reset and refractoriness: threshold should be able to return a
  single boolean, in this case reset should be given the full index array.

* Cm should be (optionally) a distributed variable (for myeline). It could be done simply by a check
that Cm is in the equations (if Cm=None, then we use the distributed variable).
* Write examples, including point processes
* Think about point processes. Use the Synapses object?
* Morphology: adding morphologies (at least cylinders)?
* One thing is missing in the Morphology: the position of the current branch in the
  big matrix (Note: all other variables, length etc, are views on the big matrix,
  for the current branch).
 * GPU
 