SQL scripts as models

SQL model steps take the form of update statements:

UPDATE input_1 SET field_1 = field_1 * 2 WHERE field_2 = "some_filter";

This injects a model into the chain whose output is the input dataset with the update operations applied. The input dataset is not modified - we use the update form because it is clearer than a select when there are a lot of columns, most of which are unchanged.

Implemented by converting into spark sql or operations on a pandas data frame, mostly used for overlays.

No files, tests, or entry points are named. Start by locating the model-step handling and the Spark SQL and pandas execution paths, then clarify how UPDATE syntax is parsed and how overlays are represented. Done means an agreed implementation with coverage for the stated non-mutating input behavior and both execution backends.

Written by the indexing model from the issue text.