Correct numerics of gif_cond_* neurons

[heplesser]

While analysing #1097, we discovered that the numerics of the gif_cond_ neuron have slight weakness: The spike triggered current is evolved outside the ODE solver, so a constant stc value is applied during all sub-steps of a NEST time step. The stc should be updated inside the rhs function, to have exact values at each substep.

[clinssen]

For more details, see #1582

[github-actions]

Issue automatically marked stale!

[heplesser]

@clinssen @pnbabu Ping!

[clinssen]

Do we prefer the precise analytic solution to be computed inside the GSL solver inner loop? This would provide numerical precision at the expense of some more exp() calls. We can also numerically approximate this decay inside the GSL loop and replace it with precise values at the end of every timestep. In the first case, we will need to dynamically allocate the ODE state vector, as the model supports an arbitrary number of STC and SFA time constants, and these will all have to go into the state vector.

[heplesser]

I would argue for precision at the expense of exp() calls.

[github-actions]

Issue automatically marked stale!