fix: fix included columns in contingnecy tables

A bug was introduced in #131, which caused columns not to be filtered if the parameter `columns` is specified resulting in useless contingency tables being generated.

To reproduce the previously existing bug:
```python
import edvart
dataset = edvart.example_datasets.dataset_titanic()

report = edvart.Report(dataset).add_bivariate_analysis(columns=['Age', 'Fare', 'Survided', 'Sex']).show()
```
The above example displays contingency tables for numeric columns `Age`, `Fare`, which are numeric and thus contain many unique, resulting in huge & useless contingency tables being displayed.

edvart/report_sections/bivariate_analysis.py

@@ -813,16 +813,22 @@ def contingency_tables(
             Data based on which to create a contingency table.
         columns : List[str], optional
             Which columns to generate pair-wise contingency tables for.
-            All columns which contain more than 1 unique value are used by default.
-            Columns which contain only null values are always excluded.
+            Columns with more than `table_threshold` unique values are excluded.
+            Columns which contain only null values are excluded.
+            To override the excluded columns, specify `columns_pairs`.
+            Ignored if `columns_x` and `columns_y` or `columns_pairs` is specified.
         columns_x : List[str], optional
             If specified, contingency tables are plotted for each pair
             in the cartesian product of `columns_x` and `columns_y`.
+            Columns with more than `table_threshold` unique values are excluded.
+            Columns which contain only null values are excluded.
             If `columns_x` is specified, then `columns_y` must also be specified.
             Ignored if `columns_pairs` is specified.
         columns_y : List[str], optional
             If specified, contingency tables are plotted for each pair
             in the cartesian product of `columns_x` and `columns_y`.
+            Columns with more than `table_threshold` unique values are excluded.
+            Columns which contain only null values are excluded.
             If `columns_y` is specified, then `columns_x` must also be specified.
             Ignored if `columns_pairs` is specified.
         columns_pairs : List[Tuple[str, str]], optional
@@ -842,18 +848,20 @@ def include_column(col: str) -> bool:
 
         if (columns_x is None) != (columns_y is None):
             raise ValueError("Either both or neither of columns_x, columns_y must be specified.")
+        if columns is None:
+            columns = list(df.columns)
+        columns = [col for col in df.columns if include_column(col)]
+
         if columns_pairs is None:
             if columns_x is None:
-                if columns is None:
-                    columns = [col for col in df.columns if include_column(col)]
                 columns_pairs = list(itertools.combinations(columns, 2))
             else:
                 columns_pairs = [
                     (col_x, col_y)
                     for (col_x, col_y) in itertools.product(columns_x, columns_y)
                     # Filter out pairs of columns which contain the same column since
                     # they make no sense in a contingency table
-                    if col_x != col_y
+                    if col_x != col_y and include_column(col_x) and include_column(col_y)
                 ]
 
         for column1, column2 in columns_pairs: