Metadata auto-detection should find primary keys of semantic sdtypes

[npatki]

Problem Description

As of SDV v1.8.0, the metadata auto-detection can identify a wide variety of sdtypes:

• statistical columns such as 'numerical'', 'datetime', 'categorical', 'boolean'
• semantic concepts such as 'email', 'phone_number', 'latitude', etc.
• structured identifiers, 'id'

However, when detecting a primary key, it only considers columns that are sdtype 'id'. In reality, semantic columns such as 'email' or 'phone_number' may also be primary keys and should be considered as possibilities.

Expected behavior

Consider the default demo dataset. The first column, 'guest_email' is the primary key. The metadata should continue detect is as sdtype 'email', and it should also mark is a primary key.

from sdv.datasets.demo import download_demo
from sdv.metadata import SingleTableMetadata

data, _ = download_demo(
    modality='single_table',
    dataset_name='fake_hotel_guests'
)

metadata = SingleTableMetadata()
metadata.detect_from_dataframe(data)

Additional context

Note that if the column was named 'guest_id' and contained random number ids, then the metadata script would correctly identify the sdtype as 'id' and it would mark it as a primary key.