Only the columns used in the lookup should be loaded in memory, not the whole lookup dataset

Otherwise, if you try to perform a lookup on a dataset containing a reasonable number of rows but a bunch of columns ... things go awfully wrong.

The obvious workaround is to get rid of the columns in the lookup dataset, but that forces the user to have 1 lookup dataset per use case of lookup ... which is, let's face it, annoying and dumb.