Allow adding a 2D image to a 3D space view

[emilk]

Based on https://discord.com/channels/1062300748202921994/1164158709103083612

Sometimes you want to show a 2D image with some 3D data next to it, like at 1:34 in this video: https://youtu.be/bKUAZ7IQcf0?si=x6VGZTB8usL90I2v&t=94

To accomplish this one should be able to do rr.log("3d/image", rr.Image(…)) + rr.log("3d/paths", rr.LineStrips3D(paths)), and then add both entities to the same 3D space view. However, this is currently not allowed:

Repro:

import numpy as np
import rerun as rr

rr.init("rerun_example_test", spawn=True)
image = np.zeros((8, 12, 3), dtype=np.uint8)
rr.log("world/image", rr.Image(image))
rr.log("world/points", rr.Points3D([[0, 0, 0], [1, 1, 1], [2, 2, 2]]))

[jleibs]

Interestingly we only recently prevented this, as of: #3822

Rather than treating images as optionally being 3D, I'm inclined to require the explicit addition of an orthographic projection:

rr.log("world/image", rr.Image(image), rr.Orthographic.IDENTITY)
rr.log("world/points", rr.Points3D([[0, 0, 0], [1, 1, 1], [2, 2, 2]]))

Which would have the added bonus of also allowing us to control the mapping between 3D and 2D coordinates in cases where pixel coordinates don't make sense for the 3d content.

[jleibs]

Github misinterpreted my comment as a resolution