BUG: figures with height > width are too tall

[neutrinoceros]

Bug report

Bug summary

related to #3631
Currently when plotting a region that happens to be broader in "height" (vertical axis) than in "width" (horizontal axis), the output figure may be very tall and not as readable as the standard square figure produced by yt.

Code for reproduction

ds = yt.load_sample("KelvinHelmholtz")

slc = yt.SlicePlot(ds, "z", "density")
slc.save("/tmp/rect_1.png")

Actual outcome

Expected outcome

I'd except a smaller figure, that would fit my screen while staying readable, on par with what I obtain after flipping axes:

vals = ds.coordinates.y_axis["z"], ds.coordinates.x_axis["z"]
ds.coordinates.x_axis["z"], ds.coordinates.y_axis["z"] = vals
ds.coordinates.x_axis[2], ds.coordinates.y_axis[2] = vals
slc = yt.SlicePlot(ds, "z", "density")
slc.save("/tmp/rect_2.png")

I think the root cause is trivially that the figure width is hardcoded while the aspect ratio is constant.
I would make more sense to hardcode the width of the largest dimension, wether it happens to be the width or the height of the figure, and it would be backwards compatible in cases where it doesn't matter.

[neutrinoceros]

After looking into the code I think it may even be considered a bug, because the following comment shows intention of implementing the behaviour I expect, not the one we actual see

yt/yt/visualization/base_plot_types.py

Lines 343 to 345 in 7567464

	def _get_best_layout(self):
	# Ensure the figure size along the long axis is always equal to _figure_size

[neutrinoceros]

I think I got half the fix with this patch:

--- a/yt/visualization/base_plot_types.py
+++ b/yt/visualization/base_plot_types.py
@@ -343,12 +343,12 @@ class ImagePlotMPL(PlotMPL):
     def _get_best_layout(self):

         # Ensure the figure size along the long axis is always equal to _figure_size
-        if is_sequence(self._figure_size):
-            x_fig_size = self._figure_size[0]
-            y_fig_size = self._figure_size[1]
-        else:
-            x_fig_size = self._figure_size
-            y_fig_size = self._figure_size / self._aspect
+        if not is_sequence(self._figure_size):
+            self._figure_size = (self._figure_size, self._figure_size / self._aspect)
+            if self._aspect < 1:
+                self._figure_size = tuple(reversed(self._figure_size))
+
+        x_fig_size, y_fig_size = self._figure_size

         if hasattr(self, "_unit_aspect"):
             y_fig_size = y_fig_size * self._unit_aspect

running the script from the OP again yields

(the axis-flipped version is unchanged)
So the elements of the figure have the desired size, but the window size is now too wide somehow.