Fixing child reference bug by removing caching during qualification

[knassre-bodo]

The bug in question arose in PyDough queries like this:

lines_1995=lines.WHERE(YEAR(order.order_date) == 1995)
lines_1996=lines.WHERE(YEAR(order.order_date) == 1996)
parts_per_year = parts.CALCULATE(sum1995=SUM(lines_1995.quantity), sum1996=SUM(lines_1996.quantity))

The problem arises during qualification: order.order_date is derived twice, both times within the context of TPCH.parts.lines, so its value is cached. However, this caching means that the second lines collection (the one that will become lines_1996) doesn't have the side effect of having a child collection for order added to its list. This causes a bug to pop up far later in the pipeline when lines_1996 references a child node that doesn't exist in order to fetch its order date. The fix is to remove the caching during the qualification step of the pipeline (confirmed to have essentially no effect on overall performance of the tests).