include only rows with equal PK to values diff stats in --json output

[vvkh]

We used to count differences across all rows for the "diffCounts" metric, even added/removed rows, for --json output format. Replaced that with extra_column_diffs value — the one we currently use for normal output.

Normal output

--json

{
  "status": "success", 
  "result": "different", 
  "model": "model.dbt_valentin_snowflake.downstream_different_from_row_diff", ..., 
  "stats": {"diffCounts": {"VALUE": 3}}},
  ...
}