[Feature] Fetch `fail_calc` metrics even for passing tests about dbt-core HOT 1 OPEN

Thanks for opening this issue and the associated PR @tbog357 ! 🤩

Agreed with @jtcohen6 in #9657 (comment) that this sounds reasonable and that:

We'll just want to do a little bit of thinking to make sure this wouldn't be a behavior change that would surprise someone upon upgrade.

So let's take a closer look at this!

Suppose we have the following project files:

models/my_model.sql

select 1 as id union all
-- select null as id union all
-- select null as id union all
select null as id

models/_models.yml

models:
  - name: my_model
    columns:
      - name: id
        tests:
          - not_null:
              config:
                severity: error
                error_if: ">2"
                warn_if: ">1"

And then we run this command:

dbt build -s my_model

The contents of target/run_results.json currently contain JSON like this:

    "results": [
        {
            "status": "pass",
            "timing": [...],
            "thread_id": "Thread-1 (worker)",
            "execution_time": 0.04432988166809082,
            "adapter_response": {
                "_message": "OK"
            },
            "message": null,
            "failures": 0,
            "unique_id": "test.my_project.not_null_large_table_id.915a6f562e",
            "compiled": true,
            "compiled_code": "\n    \n    \n\n\n\nselect id\nfrom \"db\".\"dbt_dbeatty\".\"large_table\"\nwhere id is null\n\n\n",
            "relation_name": null
        }
    ],

If your proposed change is adopted, then the only difference would be this portion (i.e., "status": "pass" would remain):

            "failures": 1,

This sounds good to me.

The case where this would affect someone is if they are using the number of failures to determine if there was a non-zero exit code or not. If they were doing this, then they would already have issues today because the warning case already reports a number that isn't 0, and a warning is still considered passing as well.

Since it seems most accurate to report the actual number of rows flagged by the test within run_results.json (irrespective if they meet the threshold or not), I'd feel good about accepting your proposal provided that we call this out in a migration guide.

from dbt-core.