.T/transpose corrupts rendering for whole-FOV CytoDataFrames using render_whole_image (no bounding box / compartment center) #226

Environment

• cytodataframe: 0.3.3
• Python: 3.11
• Installed via: pip install cytodataframe (PyPI)

Summary

Calling .T (or .transpose()) on a CytoDataFrame built from image-level (whole-FOV) data (i.e., no bounding box columns and no compartment-center columns, using display_options={"render_whole_image": True}), produces a broken render: the output is neither transposed nor does it show the embedded image. It falls back to a plain untransposed table of raw filenames.

Without .T, the same object renders correctly with real embedded images and a correct table.

Minimal reproduction

import pandas as pd
from cytodataframe import CytoDataFrame

df = pd.DataFrame({
    "FileName_OrigDNA": ["imgA.tiff", "imgB.tiff"],
    "PathName_OrigDNA": ["/path/to/images", "/path/to/images"],
    "SomeMetric": [-2.5, -2.7],
})

cdf = CytoDataFrame(df, display_options={"render_whole_image": True})

cdf.head(2)      # renders correctly: real embedded images
cdf.head(2).T    # broken: no images, and NOT actually transposed in the rendered HTML

Expected behavior

cdf.head(2).T should render a transposed table with the image still embedded in the FileName_OrigDNA row.

Actual behavior

The rendered HTML is untransposed (same orientation as before .T) and shows raw filename text instead of the image.

Root cause

In _generate_jupyter_dataframe_html (src/cytodataframe/frame.py), around the "Re-add image path columns" block:

data = (
    data.join(other=self._custom_attrs["data_image_paths"])
    if compartment_center_externally_joined or bounding_box_externally_joined
    else self.join(other=self._custom_attrs["data_image_paths"])
)

Earlier in the method, when is_transposed is True, a correctly un-transposed working copy is built:

if self._custom_attrs["is_transposed"]:
    data = pd.DataFrame(self).transpose()

But the image-path re-join's else branch joins against self (still in its transposed shape — columns are row-ids, index is the original column labels) instead of the already-corrected data. Since neither bounding_box_externally_joined nor compartment_center_externally_joined is True for whole-FOV data (there's no bounding box or compartment center at all), this else branch is exactly the one taken. So the un-transpose fix from earlier is discarded, and the join is performed on completely mismatched indices, silently producing no usable PathName_* values. Because this corrupted data is used going forward, the later if self._custom_attrs["is_transposed"]: data = data.transpose() retranspose-for-display step also ends up applied to the wrong object, which is why the final table isn't transposed either.

Suggested fix

In that branch, join against data (the transpose-corrected copy) rather than self whenever is_transposed is True, e.g.:

data = (
    data.join(other=self._custom_attrs["data_image_paths"])
    if compartment_center_externally_joined
    or bounding_box_externally_joined
    or self._custom_attrs["is_transposed"]
    else self.join(other=self._custom_attrs["data_image_paths"])
)