Columns shift in faceted visualization with sparse data

[rmouser]

Columns shift unexpectedly in faceted visualization with sparse data.

I assumed that 2 row, 2 column faceted visualization with only 3 sets of data would populate the first row and only the first column of the second row.

Instead if populated the first row, second column and the entire second row.

In my case, rows and columns of charts are to be arranged corresponding to their physical placement in the real world.

Here's a gist: https://vega.github.io/editor/#/gist/5d9c844531f8c53576b6c50914e274ef/spec.json

Mouse over the points to see expected positions.

If you add the following json to the end of data values, the charts shift to the expected positions.

      ,{
        "ID": 4,
        "Title": "Point Y-B",
        "Column": "B",
        "ColumnOrder": 1,
        "Row": "Y",
        "RowOrder": 2,
        "XValue": 4,
        "YValue": 3,
        "Size": 10
      }

I checked for duplicate issues, saw some similar, but not exactly this.

I may add some data for invisible points to work around this, unless someone can suggest a better work around.

[PBI-David]

Related to #8675?