[CT-628] Inconsistency between get_columns_in_relation method + macro

[jtcohen6]

Follow-up to dbt-labs/dbt-core#5232

Other adapters have total consistency between:

• adapter.get_columns_in_relation (adapter class method)
• get_columns_in_relation (Jinja macro)

Whereas:

In dbt-spark, the logic is written in both Python and SQL. The adapter method calls the macro, but it also does some of its own post-processing, which is missed if you just call the macro directly.

Let's aim for consistency by:

• renaming the macro that returns unprocessed data, to get_columns_in_relation_raw
• adding a macro named get_columns_in_relation that repoints to the adapter method

This is a breaking change only for those users who have been calling get_columns_in_relation, as a macro, and expecting it to return raw/unprocessed results. We can call out clearly in the changelog + release notes that this wasn't intended behavior, and users can still achieve that behavior by calling the new get_columns_in_relation_raw macro.