User-supplied fail_calc for tests

[jtcohen6]

Describe the feature

Today, the 'test' materialization hard-codes count(*) as the way to calculate failures from a test:

https://github.com/fishtown-analytics/dbt/blob/26fb58bd1b08218781b182918f5ed4dec3f735d9/core/dbt/include/global_project/macros/materializations/test.sql#L4-L7

I think that makes sense 90% of the time in the general case, but we want to give users the ability to customize the "failure calculation" if they'd like it to be something other than count(*). This is important for backward compatibility, since schema tests could previously calculate and return whatever numeric value they wanted. In the wild, this could be as simple as sum(column) instead of count(*), or it could be as complex as the dbt_utils.equality test:

select count(*) from unioned) +
        (select abs(
            (select count(*) from a_minus_b) -
            (select count(*) from b_minus_a)
            )

I'm hopeful this is quite straightforward to implement—it's just a matter of pulling in the fail_calc and templating it into the materialization.

Questions

• Should fail_calc be a test config or a test property? I lean toward property, since I think this is an essential component of the test definition and less like something that wants to be set for many different types of tests at once, e.g. from dbt_project.yml. (In a post-Set configs in schema.yml files #2401 world, this is hopefully a less meaningful distinction!)
• Could this have potentially strange interactions with % values of warn_if / error_if (Net-new test configs #3258)? Yes! I don't think we need to solve for every edge case there now.

[jtcohen6]

Update: We're going to make this a test config, following the pattern sketched out in #3258.

I don't think it makes a ton of sense for users to override the default configs set by the generic test definition, but hey, they'll be able to if they want to. I could even see this being compelling—let's say you want the unique test to calculate failure as the number of original rows containing a duplicate, rather than the length of the set of duplicate values.

[jtcohen6]

Resolved by #3392