Text columns are always converted with one-hot encoding even when excluding them from projection

[dkammer]

See example data set on cheese; even when excluding "flavor" from the projection, we get numerous categorical one-hot filters, which make it very hard to read tooltips and navigate the filters in the sidebar. Solution: we might need an option to delegate some information only to the "metadata", which is e.g. shown in the tooltip or used by the text search filter.

cheese_details_csv.csv

[dkammer]

also see #64