Include origin of cameras that are part of scene when computing view bounds

[jleibs]

What

Although we decided not to generally include the camera frustum in the view bounds. Having the origin included in the view bounds makes it much harder to totally "lose" the camera.

Consider:

import numpy as np
import rerun as rr

rr.init("rerun_example_pinhole", spawn=True)
rng = np.random.default_rng(12345)

rr.log("world/camera", rr.Pinhole(focal_length=300, width=300, height=300))
rr.log("world/points", rr.Points3D(rng.uniform(0, 1, size=[10, 3]) + [-0.5, -0.5, 1]))

Before:

After:

I still, personally find the default frustum a little small, but that's another issue:

• Improve pinhole_image_plane_distance size heuristic #3810

Checklist

• I have read and agree to Contributor Guide and the Code of Conduct
• I've included a screenshot or gif (if applicable)
• I have tested demo.rerun.io (if applicable)
• The PR title and labels are set such as to maximize their usefulness for the next release's CHANGELOG

• PR Build Summary
• Docs preview
• Examples preview
• Recent benchmark results
• Wasm size tracking

[emilk]

Including the origin makes perfect sense to me