I ran into an issue with Seaborn 0.13.0 that wasn't present in previous versions. The following code works fine in Seaborn 0.12.2 but raises numpy.core._exceptions._UFuncInputCastingError on 0.13.0.

from datetime import timedelta
import seaborn as sns
from matplotlib import pyplot as plt

data = {
    'delta': [timedelta(1), timedelta(5), timedelta(3)],
    'x': [1, 2, 3]
}

sns.scatterplot(data, x='x', y='delta')
plt.show()

The reason are numpy's default casting rules and the fact that the check changed from np.isin(vector, [0, 1, np.nan] to np.isin(vector, [0, 1]) in the latest release. That means that previously the comparison was with 0.0 and 1.0 as float dtype, because the nan made NumPy cast the whole test array to float. Now the comparison is against int dtype.

In my opinion, the actual dtypes are not relevant here. But we should not assume that the default casting rules allow that this comparison is possible. Simply because the data type of vector is not know as the data is provided by the user.
Therefore, I'd say it is best to handle the _UFuncInputCastingError (which is a subclass of TypeError) and in that case decide that the data is not binary/boolean. The then remaining checks should not have similar issues.