spread_all() fails if JSON has two different obvious types (numeric/character) if the more restrictive type comes first.

[captbrando]

Been struggling with Azure activity audit data and two operation types, ListViewed and ListItemDeleted, return Character arrays, while everything else is either a numeric/double or NA. This causes a failure when running spread_all():

Error in `z[, final_columns, drop = FALSE]`:
! Can't subset columns that don't exist.
✖ Column `ListBaseType` doesn't exist.

The way to fix this was to split those records out into a separate data frame, change the data type for the more restrictive to more permissive (double->character), and then go from there. spread_all() should pick the less restrictive type when required.