Right now there is a normalization step which converts columnar data (aka map of seq)
This should be avoided since this can get slow for large columnar data sets
[ ] Stats in header should be able to deal with columns directly rather than assuming seq of maps
[ ] note that compute-table-summary transposes the rows again
Right now there is a normalization step which converts columnar data (aka map of seq) This should be avoided since this can get slow for large columnar data sets