Distinguish points, spheres and circles

[Wumpf]

Today, log_points takes a world/scene size radius. For 2D this makes it effectively circles with a pixel size and for 3D they become (due to the way re_renderer accounts for projection) spheres. Hidden, there is also a way to have them be ui sized: If a point doesn't have a radius it will take the global default which can be configured to be in ui points.

Proposing to instead define three distinct types:

• points
  • may have a 2D or 3D position
  • radius in viewer ui units
    • call it stroke_width or similar?
  • their radius may (for time being or indefinitely?) be completely unaffected by camera distortion, making them easy to solve for the vertex approach described in Add support for Brown-Conrady Lens distortion #2499
  • when shown in 3D, do not behave like circles (i.e. re_renderer sets FLAG_DRAW_AS_CIRCLES)
• spheres
  • always have a 3D position!
  • radius in world units
• circles
  • may have a 2D or 3D position
  • radius in world units. The expectation is that this is usually pixels on an image
  • when shown in 3D, do not behave like circles (i.e. re_renderer sets FLAG_DRAW_AS_CIRCLES)

Note that re_renderer's PointCloudRenderer (to be renamed?) can already serve all three types out of the box!
(Detail: Spheres do not yet cut correctly through other geometry, something that can be easily fixed)

3D spheres with ui size are explicitly left out.

Supersedes #2029

[Wumpf]

We have spheres/ellipsoids by now. narrowing the problem to circles & points
Still an open question whether points should do depth manipulation as proposed in this PR. Probably they should, at least when reaching a certain relative size!

• Render points with per-fragment depth, improving visuals for points intersecting with other geometry #6508