Label outlines cause wrong transformation

[aeisenbarth]

It seems outline=True takes a code path that causes labels to have a wrong scale.

It is important for the images that I render to not have filled labels, but outlines.

Example

import numpy as np
import spatialdata as sd
import spatialdata_plot # noqa: F401
from skimage.transform import AffineTransform

# Same array, same transformation used for both image and labels.
# They should exactly overlap.
array = np.zeros((100, 100), dtype=int)
array[30:70, 30:70] = 1
affine = sd.transformations.Affine(
    AffineTransform(translation=(10, 10), rotation=np.deg2rad(15), scale=0.65).params,
    input_axes=("y", "x"),
    output_axes=("y", "x"),
)
image1 = sd.models.Image2DModel.parse(
    array[:, :, np.newaxis], dims=("y", "x", "c"), transformations={"global": affine}
)
labels1 = sd.models.Labels2DModel.parse(
    array, dims=("y", "x"), transformations={"global": affine}
)
sdata = sd.SpatialData(images={"image1": image1}, labels={"labels1": labels1})

# Render with defaults. Spatialdata-plot fills the labels with color, but no outline.
sdata.pl.render_images("image1").pl.render_labels("labels1").pl.show(title="without outline", save="without_outline.png")

# Render without fill, but with outline, because I don't want to occlude what is behind the labels.
sdata.pl.render_images("image1").pl.render_labels(
    "labels1", fill_alpha=0.0, outline=True, outline_alpha=1.0
).pl.show(title="with outline", save="with_outline.png")

[timtreis]

@aeisenbarth I think we fixed the issue as a side-effect of the last PRs - at least I cannot reproduce it anymore. Could you verify?

If it's solved, could you remove your changes to render.py from #275? I'd be happy to merge in your test and the resulting image

[aeisenbarth]

Yes, the issue was also fixed by the refactoring of _render_labels. Thanks!
I rebased the branch to only include the test.