After a lot of experimentation I was able to do this with ADF pipeline but don't recommend this since it is easy to miss fields in this approach and it works only if the schema is fixed. It basically works by bringing the nested field to the root, updating it and then joining it with the rest of the data.

Step 1: Create two branches for the input data

Branch 1:

1. Select: Select properties.execution AS execution, OrderID

2. Select: Select all properties in execution: Select execution.item AS item, OrderID

3. Derived column: items = Array(item)

4. Construct execution object - Derived column with subcolumns item, items

5. Select execution, OrderID

Branch 2:

1. Select all properties except properties.execution

Join: Branch1, Branch2 on OrderID

Dervied column: construct properties with subcolumns execution, and other fields within properties

Select: finally select only the required fields and output