[Metadata Detection] Only make primary/foreign keys sdtype id (leave others as unknown)

[npatki]

This issue is for the in-progress metadata-improvment branch, which aims to improve the accuracy of metadata detection from datasets.

Problem

The new metadata detection logic is described in #1515. This is implemented according to spec but we've found that it's not intuitive.

When the values in a column are unique, the script is very liberal in calling it an 'id' column. In reality, many PII columns will also have unique values (eg. SSNs, phone numbers, email addresses, etc.).

Expected Behavior

To not automatically assume that all unique values are sdtype 'id'. The end result should be the following:

• Primary keys should be sdtype 'id'
• Foreign keys should be sdtype 'id'
• Other columns with unique values should be sdtype 'unkown' (so they are treated as PII)

Details

This may be accomplished by a few changes to the logic in #1515:

• 3 (iv): Elif the values are all unique, sdtype is id. unknown. Add it to a list of potential primary keys.
• 5 (iv): Elif the values are all unique, sdtype is id unknown. Add it to a list of potential primary keys.

Then when detecting relationships:

• Primary key: We can continue to choose the leftmost column of the potential keys we've identified. Only that column should have sdtype updated to 'id'. (See Detect primary keys in metadata #1521 for initial specs)
• Foreign key: No changes should be needed (See Detect relationships in MultiTableMetadata #1522)