Referential integrity output needs to be less verbose and more informative

[vijaykiran]

Given this SodaCL

checks for fact_product_inventory:
  - reference from (product_key) to dim_product (product_key)

the output of running soda scan results in 111455 lines because there are too many rows in fact_product_inventory with a product_key that's not there in dim_product. This makes it very inconvenient to see which keys are missing.

...
483         20140616 2014-06-16    '$36.80'  0        0         4            
483         20140617 2014-06-17    '$36.62'  0        0         4            
483         20140618 2014-06-18    '$36.78'  0        0         4            
483         20140619 2014-06-19    '$36.88'  0        0         4            
483         20140620 2014-06-20    '$36.86'  0        0         4            
483         20140621 2014-06-21    '$36.84'  0        0         4            
483         20140622 2014-06-22    '$37.01'  0        0         4            
483         20140623 2014-06-23    '$37.13'  0        0         4            
483         20140624 2014-06-24    '$37.29'  0        0         4            
483         20140625 2014-06-25    '$37.19'  0        0         4            
483         20140626 2014-06-26    '$37.13'  0        0         4            
483         20140627 2014-06-27    '$37.13'  0        0         4            
483         20140628 2014-06-28    '$37.26'  0        0         4            
483         20140629 2014-06-29    '$37.16'  0        0         4            
483         20140630 2014-06-30    '$37.22'  0        0         4   
...

Suggestion: only show the distinct columns (in this case product_key) that are violating the referential integrity check.

Context: https://soda-community.slack.com/archives/C032BDZN3HV/p1645463084048419

[tombaeyens]

This will be tackled by #1237 Samples upload to Soda Cloud.