Support scale_firing_rates with Regular/Poisson/Stochastic spiking wrappers

[arvoelke]

According to the documentation:

nengo-dl/nengo_dl/converter.py

Lines 69 to 79 in e9b359a

	scale_firing_rates: float or dict
	Scales the inputs of neurons by ``x``, and the outputs by ``1/x``.
	The idea is that this parameter can be used to increase the firing rates of
	spiking neurons (by scaling the input), without affecting the overall output
	(because the output spikes are being scaled back down). Note that this is only
	strictly true for neuron types with linear activation functions (e.g. ReLU).
	Nonlinear neuron types (e.g. LIF) will be skewed by this linear scaling on the
	input/output. ``scale_firing_rates`` can be specified as a float, which will
	be applied to all layers in the model, or as a dictionary mapping Keras model
	layers to a scale factor, allowing different scale factors to be applied to
	different layers.

Nonlinear neuron types (e.g. LIF) will be skewed by this linear scaling on the input/output.

But this does not need to be the case. For instance, when using any of the spiking wrappers in Nengo v3.1.0, e.g., RegularSpiking, StochasticSpiking, or PoissonSpiking, as a target for the conversion, the rate can be scaled correctly by multiplying dt by scale_firing_rates (https://arxiv.org/abs/2002.03553). It might make sense to have this as a configurable parameter on these NeuronType wrappers in core Nengo. Then this rescaling would be supported by any backends that support these particular wrappers.

More generally, this strategy of scaling dt works for any target neuron model, even nengo.LIF (see nengo/nengo#1427) but supporting this for the spiking wrappers would be the low-hanging fruit. It's also not obvious what backends would be able to support this more general kind of rescaling (e.g., SpiNNaker probably could, while Loihi could not).