[SL-1987] Post-join filter application on time spine queries results in invalid query exceptions about metricflow HOT 3 OPEN

To be clear, the things we need for a complete solution are:

1. Remove the existing double-filter logic
2. Enable pushdown of metric_time or agg_time_dimension filters to the time_spine

This issue should be closed when the first of these is complete, and predicate pushdown will account for the time spine in a more natural way. I've linked it to that main tracking task so we don't forget to push to time spine, but that's hardly a thing I'm likely to forget given the layout of the code.

from metricflow.

@tomkit.lento I don't think we want to remove the double-filter logic, do we? My thought is that we'll end up with the scenario where you've selected a categorical dimension, filtered out nulls, and then we join them back in so you see nulls in the output. I think the customer would expect the where filter to override the join_to_timespine directive.

My proposed solution is to only apply the second filter if the filtered spec is also requested in the group by (since those should be the only ones that might return unexpected results).

from metricflow.