Fix crash on int-coded label columns in ridgeline plot

[alicia01101]

Problem

• Fixes a crash in GeneralVisualizer._plot_latent_ridge when coloring a Ridgeline plot by an integer-coded categorical label column (e.g. class labels) with more than 3 unique values.
• Cause of crash: the function used isinstance(x, float) to decide which label values were valid numbers before quantile-binning. int values fail that check and were coerced to NaN. Problem: When all values in a label column are integers (which is the case for example for the MNIST dataset mapping labels), the entire series became NaN and pd.qcut raised ValueError: Bin edges must be unique because every computed bin edge was NaN.
• Fix suggested in this PR: label columns that are only integer-valued (no floats mixed in) are now treated as categorical directly, skipping the quantile-binning branch entirely. Columns that do contain genuine float values keep the
  existing quantile-binning behavior and non-numeric outliers within such columns are still coerced to NaN like before.
• this means that there is a behavior change for float columns which have int values mixed into them then these values are no longer dropped to NaN.

Limitations

• the crash hasn't happend before since every other tutorial currently colors by a string-typed column (eg. cancer_type, cell_type, disease...)

• however possible problem with this fix could be: a column that is integer-valued but semantically continuous (like age in years without missing values) will now be treated as categorical rather than quantile-binned

• Instead of this handling it would also be possible to simply extend the numeric check to include int, not just float, to prevent the crash

[jan-forest]

I think the root error was the code line converting valid integer values into nan-values, which does not make sense and the comment above looks like a co-production of AI and myself:

                # Change all non-float labels to NaN
                labels = [x if isinstance(x, float) else float("nan") for x in labels]

Was introduced here: bd02297

To keep the original behaviour regarding integer-based annotations like age, I suggest the following fix:

                # Try to convert all labels to float, if fails, convert to nan
                for i in range(len(labels)):
                    try:
                        labels[i] = float(labels[i])
                    except ValueError:
                        labels[i] = float("nan")
                # Check if all labels are NaN, convert to string
                if all(np.isnan(labels)):
                    labels = [str(x) for x in labels]
                else:
                    labels = list(
                        pd.qcut(
                            x=pd.Series(labels),
                            q=4,
                        labels=["1stQ", "2ndQ", "3rdQ", "4thQ"],
                    ).astype(str)
                )

[jan-forest]

We need to be careful the fix would be necessary at multiple locations in general visualizer and xmodalix visualizer

[jan-forest]

@alicia01101 I have introduced my suggested fix. Please, test with the tutorials, then I will merge.

[alicia01101]

The Tutorials work with the suggested fix!