Description

Attempt to fix #4768 and fix #4881 without bringing back the issue from #4663.

Reproducer for #4663:

q = Quaternionf(0.6849482f0, 0.175505f0, 0.15756221f0, 0.6893606f0)
f,a,p = scatter(Point3f(0), rotation=q, markersize=20, marker=Makie.logo())
scatter!(
    1:10, 1:10, rand(10, 10) .* 10,
    rotation=normalize.(rand(Quaternionf, 10 * 10)),
    markersize=20,
)
f

This produces Float32[17.774902 -0.46377563; -9.153168 0.24081421] as the marker transformation matrix. Viewing this as a basis transformation gives two basis vectors that are very close to linearly dependent:

julia> M[:, 1] ./ norm(M[:, 1])
2-element Vector{Float32}:
  0.88904786
 -0.4578143

julia> M[:, 2] ./ norm(M[:, 2])
2-element Vector{Float32}:
 -0.88749003
  0.46082675

I believe that's the issue for Cairo.

Type of change

• Bug fix (non-breaking change which fixes an issue)

Checklist

• Added an entry in CHANGELOG.md (for new features and breaking changes)
• Added or changed relevant sections in the documentation
• Added unit tests for new algorithms, conversion methods, etc.
• Added reference image tests for new plotting functions, recipes, visual options, etc.