Update Referential Integrity metric to support NaNs in child column

[gsheni]

Problem Description

• For the Referential Integrity metric, we currently do the following:
  • If the foreign key in the real data has any NaN values, those are dropped in the foreign key in the synthetic data.
  • We also verify that all foreign keys have valid parent.
• To support NaNs values in foreign keys, we need to update this logic.

Expected behavior

• Update ReferentialIntegrity.compute_breakdown() to do the following:
  • If the real data has NaN values, and the synthetic data has NaN values in the FK, then it is okay.
  • If the real data does not have NaN values, but the synthetic data has NaN values in the FK, this will result in a lower score.
  • The score can be lowered by the number of extra NaNs present in the synthetic data.
• Update test_referential_interity.py
• Update documentation

Additional context

• Function to update: https://github.com/sdv-dev/SDMetrics/blob/51f3ac2904c065462097c80f4387b973657ad6d9/sdmetrics/column_pairs/statistical/referential_integrity.py#L36