This is the bug I started the day with. And I kept finding other issues on the way. I think it's now fixed, but this made me realise how convoluted this exporter is :(

Blender Exporter: I don't think parsing the groups is needed. It exports the objects twice (or more) otherwise.

This is a formatting commit, it removes an unnecessary else branch.

The functionality removed is repeated directly after this code block.
The entire process is actually being done twice, which is undoubtedly incorrect or otherwise very unclear.

Given that it is simply doing 3 actions:

a) Adjusting the pitch/yaw based on the mouse position deltas, then multiplying that by the lookSpeed.
b) Clamping, transforming the pitch space (to the X,Z coordinate frame) and then converting the pitch/yaw (lat/lon) to radians.
c) Calculating the target position 100 units in the direction specified by the pitch/yaw.

This process has no reason to be repeated.

Plane: Removed normalize() from setFromNormalAndCoplanarPoint and added it back to setFromCoplanarPoints. As suggested in 5c6b5746.

CanvasRenderer+SVGRenderer: Removed 2D clipping for Face3/Face4 elements as the projector is already taking care of that.

However, this can create glitches on extreme cases when the polygon is inside the frustum but the vertices are not.
Maybe this can be achieved with Box3 in a performant way?
/ping @bhouston

Change the docs to reflect the true default of the fog parameter in MeshBasicMaterial.