Speed up Snowflake column comments, while still avoiding errors

[jtcohen6]

resolves #3541, slack thread
see also: #3149, #3039

Description

#3149 did a wonderful job of resolving the undesirable behavior (originally reported in #3039) whereby a column description, specified in yaml, would cause a model to fail building if the column doesn't actually exist in the model query. This "strictness" could be a feature, someday, but it's not how we think about properties (and especially descriptions) today.

Unfortunately, the approach taken in #3149 is resulting in much, much slower builds for models with many description-bearing columns—from seconds to minutes. This is because the Snowflake python connector requires running each semicolon-delineated query on its own, and so they run in sequence, hundreds of times. (Whereas dbt-postgres, which also uses per-column comment on statements, can run big batches of semicolon-delineated statements all at once.)

This PR takes an alternative approach—option 2 that I outlined in the original issue—by checking to see which columns exist in the just-created table, via adapter.get_columns_in_relation, and then comparing the result against the dictionary of columns with descriptions defined.

I'd like to sneak this into v0.20.0 final if possible :)

Checklist

• I have signed the CLA
• I have run this code in development and it appears to resolve the stated issue
• This PR includes tests, or tests are not required/relevant for this PR — Existing tests, including the one added by Check if a snowflake column exists before altering its comment #3149
• I have updated the CHANGELOG.md and added information about my change to the "dbt next" section.