performance problem: various table operations

[WinPlay02]

Describe the bug

Some Table operations take multiple hours to complete on large tables, although they should be much faster. When using pandas loc functionality, these operations complete in a few seconds.
These operations include:

• filter_rows
• group_rows_by
• add_rows(Table)
• sort_rows (probably, untested)
• transform_column (probably, untested)

The common problem of these operations are, that the first step is boxing the rows to a pandas Series and wrapping them in a Row object (which both leads to a huge overhead).
Most of the time, the single row objects are only needed to be passed to the lambda function that is provided to these operations.

Possible resolutions:

• don't guarantee that a Row is passed to the lambda functions (for filter_rows, group_rows_by, sort_rows, transform_column). Instead only pass a general type that is indexable with Strings (column names), like Mapping[str, Any] to access the values. Maybe also pass the schema, as the other needed operations on a Row object can be handled with a schema.
• add_rows could handle adding a whole table as a special case, so they are not split into Rows first, avoiding the problem without changing the interface

To Reproduce

1. Load a huge dataset with Table.from_csv_file
2. Run any of the previously mentioned operations
3. Observe that nothing happens and after a few hours the pipeline still runs

Expected behavior

Faster execution of the listed operations

Screenshots (optional)

No response

Additional Context (optional)

No response

[WinPlay02]

Concrete solution for Table.add_rows (insert at the top of the method):

        if isinstance(rows, Table):
            if rows.number_of_rows == 0:
                return self
            if self.number_of_columns == 0:
                return rows
            different_column_names = set()
            different_column_names.update(set(self.column_names) - set(rows.column_names))
            if len(different_column_names) > 0:
                raise UnknownColumnNameError(
                    sorted(
                        different_column_names,
                        key={val: ix for ix, val in enumerate(self.column_names)}.__getitem__,
                    ),
                )

            new_df = pd.concat([self._data, rows._data]).infer_objects()
            new_df.columns = self.column_names
            schema = Schema.merge_multiple_schemas([self.schema, rows.schema])
            result = Table._from_pandas_dataframe(new_df, schema)

            return result

[lars-reimann]

Fixed in #734 and #738.