shape parameter convention in layers vs preprocessing

[FabianSchubert]

The connect method of Conv2D assumes that the shape parameter of the input layer represents (height, width, channels):

ml_genn/ml_genn/ml_genn/connectivity/conv_2d.py

Line 34 in 8822f90

conv_ih, conv_iw, conv_ic = source.shape

However, the preprocess_tonic_spikes function seems to follow a convention where shape[0] is the width of the input and shape[1] is the height, given that 2d event coordinates are provided:

ml_genn/ml_genn/ml_genn/utils/data.py

Lines 99 to 100 in 8822f90

	if ("x" in ordering) and ("y" in ordering):
	spike_event_ids = events["x"] + (events["y"] * shape[0])

This is not a bug per se, since the latter is just a preprocessing utility, but I think adhering to a consistent convention for the order of width and height would help avoiding errors down the line when building a preprocessing/training pipeline.

Edit: Overall I would prefer the convention of representing it as (height, width), as this would follow the usual matrix representation for image data.