Warn users to save their metadata file after auto-detecting/updating it

[npatki]

Problem Description

As specified in the Metadata docs the metadata auto-detection logic is not meant to be accurate or complete. Furthermore, the metadata auto-detection logic may change in between SDV versions, leading to inconsistent results.

For eg. the following script may not produce the same results in every SDV version because the auto-detection script changes!

from sdv.metadata import MultiTableMetadata
from sdv.multi_table import HMASynthesizer

metadata = MultiTableMetadata()
metadata.detect_from_dataframes(my_data) # this logic is not guaranteed to be accurate and may change!!

synthesizer = HMASynthesizer(metadata)
synthesizer.fit(data)

To avoid these issues, the SDV team strongly recommends saving the metadata as a separate JSON file. This is not communicated to the user strongly enough, and leads to confusion.

metadata.save_to_json('my_metadata.json')

Expected behavior

When initializing a synthesizer, we should warn the user if they are providing a metadata object that has been auto-detected/modified but has never been saved.

 # any single, multi or sequential synthesizer
synthesizer = GaussianCopulaSynthesizer(metadata)

Warning: We strongly recommend saving the metadata using 'save_to_json' for replicability in future SDV versions.

Additional context

The warning does not need to show up if:

• The user has called the save_to_json() function on the metadata (meaning that they are following the recommendation) OR
• The user has created the metadata using load_from_json() (meaning that they are loading a previously-saved version of it)
• The user retrieved the metadata object from our download_demo() function

For all of the above: The warning should reappear if a user update the metadata afterwards using the Python API, or if they call auto-detect on it.

One way to accomplish this would be by setting/unsetting a private flag within the metadata object itself.