Reimplement ncol3 requirement

[cynthiahqy]

• add ncol3 check to new_xmap_df()
• add column dropping message to as_xmap()
• fix xmap_reverse() tests
• add df_check_ncol3() to validator fns

I'm not sure offering helpers for modifying crossmaps is necessary or even desirable.

• Lists of maps make more sense from a "using xmaps" perspective,
• Combining multiple crossmaps into a single table add a flexible quality to xmaps that goes against the purpose of the structure as a data-provenance object.
• Users can still manipulate and store multiple sets of weights or other xmap variants in a "parent table", from which they create multiple xmap.
• Manipulation of xmaps creates risk around invalidating the map -- i.e. modifying values in the weights column is technically possible for an xmap (since it is just a data.frame column), but not desirable since the map is not revalidated after the manipulation -- i.e. xmap |> dplyr::mutate() should probably return a tibble rather than an xmap.

Originally posted by @cynthiahqy in https://github.com/cynthiahqy/conformr-project/issues/67#issuecomment-1436365643

[cynthiahqy]

ncol3 created some issues with designing visualisation functions -- i.e. it makes sense to keep the long names together with the xmap?

[cynthiahqy]

Less strict solution to desire for immutability:

• Switched default .keep_all arg in as_xmap() to FALSE
• Remove xmap_tbl class,

[cynthiahqy]

Continued in #98