Better error messaging for nullable foreign keys

[npatki]

Problem Description

Currently, the SDV does not support nullable foreign keys though this feature request has been filed in #1656. We expect support for nullable foreign keys to take some time, as it requires core algorithmic updates to each multi-table algorithm.

If you have a null foreign key, it currently shows up an an error. For eg. in demo dataset TubePricing_v1:

Data does not match metadata for dataset TubePricing_v1 The provided data does not match the metadata:
Relationships:
Error: foreign key column 'connection_type_id_2' contains unknown references: (nan). All the values in this column must reference a primary key.
Error: foreign key column 'connection_type_id_1' contains unknown references: (nan). All the values in this column must reference a primary key.

This message misleading because the data is actually valid -- databases do support nullable foreign keys. We should clear up confusion that users may have about this missing feature.

Expected behavior

• The metadata.validate_data call should allow nullable foreign keys to exist. That is: It should not crash.
• Each multi table synthesizer should instead make this call and clearly communicate to the user that the feature is unsupported. In the future, we will add support for each synthesizer individually, in which case that particular synthesizer be updated.

>>> metadata.validate()
>>> metadata.validate_data(data)
>>> synthesizer = HSASynthesizer(metadata)
>>> synthesizer.fit(data)
SynthesizerProcessingError: The data contains null values in foreign key columns. This feature is currently unsupported. Please remove null values to fit the synthesizer.

Affected columns:
Table 'transactions', column 'session_id'
Table 'sessions', column 'user_id'