Revise input and output ports in NESTML

[clinssen]

Introduces the NESTML predefined sift() function to process spike events in event handlers. See here for documentation on sift(): https://github.com/nest/nestml/pull/1299/files#diff-22c1d3766680291b0aed3cd548fc0980ecc6511ebb5532a8ed2607105e1775e3R1198

Remove the excitatory and inhibitory spiking input port qualifiers. Instead, spikes should be routed (a) on the basis of having multiple spiking input ports (like one for excitatory and one for inhibitory input); or (b) on the basis of a condition like

if weight < 0: 
    # treat as an inhibitory spike
else:
    # treat as an excitatory spike

A consequence of this is that spike ports can no longer be assumed to be linear and time-invariant in the general case (as in this case, a spike from the buffer will be processed differently depending on the sign of its weight). On the C++ side, this means that each spiking input port now uses the VectorRingBuffer class (https://github.com/nest/nestml/pull/1299/files#diff-ee49404cf3290ce0e9ee40e25068b6d123da357aee83794ab89745a7e2aaf988R1) rather than the RingBuffer. The performance penalty of this is approximately 12.5% in a small balanced network example. An extra code generation option, linear_time_invariant_spiking_input_ports, is provided to instruct the NEST code generator that a set of given ports can be considered LTI. With this option enabled, performance is indistinguishable from the current main branch performance.

Should be merged after #1288. Done.
Should be merged after #1349. Done.
Should be merged after #1358. Done.
Should be merged after #1291.

Replaces #1137.

Fixes #944.

[tomtetzlaff]

Suggested change

	Typically, the statements in the ``onReceive`` block integrate the delta function across time, yielding the surface area under the curve, the value of which corresponds to the weight of the spike. To perform this integration in a concise way, we use the sifting proprty of the delta function. The sifting property states that for any continuous function :math:`f(t)`:
	Typically, the statements in the ``onReceive`` block integrate the delta function across time, yielding the surface area under the curve, the value of which corresponds to the weight of the spike. To perform this integration in a concise way, we use the sifting property of the delta function. The sifting property states that for any continuous function :math:`f(t)`:

[tomtetzlaff]

Very nice. Apart from the small typo, this looks very good to me. Thanks a lot for this nice solution.

[clinssen]

Thank you very much for the review and the kind thumbs up!