Unify neuron + synapse as hybrid dynamical systems models; clarify integration loop sequence

[clinssen]

Fixes #440
Fixes #192
Fixes #938
Fixes #189
Fixes #598

For the rendered documentation, see https://nestml-sandbox.readthedocs.io/

• add feature to integrate specific ODEs with integrate_odes()
• convolutions are always integrated in time, independent of whether integrate_odes() is called, and in the case of NEST Simulator, are always assumed to be processed after the subthreshold dynamics but before the condition checks; hence the observed value will always be the one "just after" the instantaneous change due to the spike event
• integration sequence (what steps take place in one timestep update) clarified
• "neuron" and "synapse" top level elements merged; now just one "model". Whether to generate code for a model as a neuron or as a synapse is determined only during code generation.
• added "onReceive" event handler blocks for spiking input ports. Priorities can be specified for event handlers in case of events occurring at the exact same time.
• added "onCondition" blocks to check conditions

To do

• merge emit_spike() and deliver_spike(). How to deal with parameters of the spike event itself? Spun off to Adding attributes to spike events #946.
• generate code for onEvent blocks in neurons
• rewrite neuron models (split update block into update, onCondition blocks)
  • fix nest_integration_test.py
  • mat2_psc_exp_neuron, izhikevich_psc_alpha_neuron, iaf_psc_exp_htum_neuron, iaf_psc_exp_dend, iaf_cond_exp_sfa_rr_neuron, iaf_chxk_2008_neuron, hill_tononi_neuron, hh_moto_5ht_neuron, hh_cond_exp_traub_neuron, hh_cond_exp_destexhe_neuron, doc/tutorials/spike_frequency_adaptation/models/*.nestml
  • ignore_and_fire_neuron
• implement priority attribute for event handlers
• come up with a more efficient way of integrating only a subset of the ODEs in case numerical solver is used
  • Python code generator
  • NEST code generator
• Fix Jupyter notebooks
• Manually test on SpiNNaker
• Fix compartmental target (simply disallow calling integrate_odes() with arguments for now)
• Allow for it to be specified as a code generator option which of the models are to be rendered as neurons and which as synapses (right now it's clumsily derived from the model name suffix)
• Double-check coco behaviour: is a warning emitted for parameters/internals/state without a defined initial value? Spun off to Emit warning for uninitialised state/parameter/internal variables #980
• Support priorities for onCondition blocks -- Spun off to Support priorities for onCondition blocks #981
• look into/fix the line // B_.spike_input_received_[{{ inport.upper() }} - MIN_SPIKE_RECEPTOR] = false; in NeuronClass.jinja2
• look into/add docs for the line {# for neuron-synapse co-generation: separate integrator for the emulated/dummy integrate_odes() below in this template #} in NeuronClass.jinja2
• Fix Python 3.8 compatibility (strings removesuffix())
• Documentation
  • main integration loop figure
  • integration_strategies figure
  • re-generate RST/HTML documentation for all the models
    • merge autodoc code generator and extras/nestml_model_doc_generator.py -- maybe something in resouces_autodoc?

[clinssen]

Should be merged after #897. Please wait with your review until this PR is out of draft state. Cheers! Edit: done!

[clinssen]

Thanks for the heroic review and comments! The remainder of the comments will be addressed in #1025 to keep the size and reviewability of this PR manageable.