Add attributes to spike events

[clinssen]

Fixes #946.
Fixes #944.

Adds attributes to input spike events (see documentation updates inside this PR for more details).

The spike event handle() method now needs to store each received spike in a separate buffer entry, otherwise excitatory and inhibitory spikes (which are distinguished by sign of the weight); if they are not in a separate buffer, an inhibitory spike with weight -4 and excitatory spike of weight 1 would end up as an inhibitory spike of -3). A consequence of this is more buffering is needed, which could result in a performance hit. A code generator option should be introduced, that gives a list of spiking input port names, such that for each port, it allows the regular RingBuffers to be used.

• The processing of spikes in the FIR filter is now delayed by one resolution timestep. The impact and/or mitigations of this should still be discussed.
• A new code generator option should be added that allows two separate exc/inh input ports to be merged into one just like was the case originally for NEST Simulator, replicating the existing behaviour for a single rport-model that routes spikes depending on the sign of the weight.
• A new syntax should provide a difference between grabbing the instantaneous event attribute (for instance, a weight in pA), and adding series of delta pulses to an ODE (where units are multiplied by 1/s to achieve, for instance, pA/s). The proposed syntax for this is to use the attribute as an explicit function of time in the ODEs by appending the (t) suffix, for instance: x' = -x / tau + spikes_in_port.weight(t) (in the equations block) versus x += spikes_in_port.weight (in the onReceive block). Or we require the attributes to be convolved with a delta kernel in the onReceive blocks:
  onReceive(spike_in_port): I_syn += convolve(delta(t), spike_in_port.weight)

Benchmarking shows that performance degradation for the most general case (without linear_time_invariant_spiking_input_ports enabled) is only on the order of 12.5%. With linear_time_invariant_spiking_input_ports enabled, performance is indistinguishable from what it was before this PR.

Notes for reviewers

See especially the following sections in the documentation to review:
From

https://nestml-sandbox.readthedocs.io/en/latest/nestml_language/nestml_language_concepts.html#spiking-input-ports

until (not including) the section "Guards".