Incorrect column name ordering for Multi-Table Synthesizer

[R-Palazzo]

Environment Details

• SDV version: 1.17.1

Error Description

When generating synthetic data, we consider the metadata order as the source of ground truth for the column names.
However, because of this line in the BaseMultiTableSynthesizer, we use the original column order of the real data currently:

SDV/sdv/multi_table/base.py

Lines 526 to 529 in 315266f

	table_columns = getattr(self, '_original_table_columns', {})
	for table in sampled_data:
	if table in table_columns:
	sampled_data[table].columns = table_columns[table]

This creates inconsistency between the single-table and multi-table behaviors

Steps to reproduce

Here is a code for HMA

import pandas as pd
from sdv.metadata import Metadata
from sdv.multi_table import HMASynthesizer

table_1 = pd.DataFrame({
    'col_1': [1, 2, 3],
    'col_3': [7, 8, 9],
    'col_2': [4, 5, 6],
})
table_2 = pd.DataFrame({
    'col_A': ['a', 'b', 'c'],
    'col_B': ['d', 'e', 'f'],
    'col_C': ['g', 'h', 'i'],
})
metadata = Metadata.load_from_dict({
    'tables': {
        'table_1': {
            'columns': {
                'col_1': {'sdtype': 'numerical'},
                'col_2': {'sdtype': 'numerical'},
                'col_3': {'sdtype': 'numerical'},
            },
        },
        'table_2': {
            'columns': {
                'col_A': {'sdtype': 'categorical'},
                'col_B': {'sdtype': 'categorical'},
                'col_C': {'sdtype': 'categorical'},
            },
        },
    }
})
data = {
    'table_1': table_1,
    'table_2': table_2,
}

synthesizer = HMASynthesizer(metadata)
synthesizer.fit(data)
synthetic_data = synthesizer.sample()
synthetic_data['table_1']

The printed output is:

While using GaussianCopula:

from sdv.single_table import GaussianCopulaSynthesizer

metadata_single_table = Metadata.load_from_dict({
    'columns': {
        'col_1': {'sdtype': 'numerical'},
        'col_2': {'sdtype': 'numerical'},
        'col_3': {'sdtype': 'numerical'},
    }
})
synthesizer_single_table = GaussianCopulaSynthesizer(metadata_single_table)
synthesizer_single_table.fit(table_1)
synthetic_data_single_table = synthesizer_single_table.sample(num_rows=3)
synthetic_data_single_table

The printed output is: