Description
When we select a column from the dataframe view, we previously made the decision that a missing column would provide results but be implicitly null.
The rationale for this decision was that in many situations, Rerun data producers may not even create a column if something is missing. For example: maybe a detector simply never logs anything if nothing is detected. In these cases, a query-writer reasonably wants a null column to exist.
However, the far, far far more common case is someone exploring the data and writing a test script made a typo. If a user doesn't actually know what was in the dataset, they may think this is validly null data, as opposed to realizing they aren't querying the right content.
We should instead default to error on null, and then provide a mechanism for advanced users to inject any missing-but-required columns into the view only in situations where it makes sense to do so.