Update collapse_col() / collapse_col_flex()

[allenbaron]

• Add ignore_na parameter
  • Should work for all methods
  • unique already has na.rm parameter --> implement it
  • write wrapper function around first and last
    1. use na.omit
    2. if zero length = return NA of same type, otherwise use first or last

• Consider implementing ignore_na in _flex() function to account for differential handling of methods
  • boolean = apply to all or none
  • vector (character or named) = apply to specified columns

• Consider swapping {{ .cols }} with dplyr::everything() in summarize
  • To allow for negative selection to work, e.g. c(-x, -y) --> I want to collapse all columns except x & y

[allenbaron]

It would be nice to make this as flexible as dplyr verbs. These functions are essentially just a group_by() + summarize() operation, so perhaps they should be consolidated & only collapse_col() kept.

The could be done by removing the default method parameter and requiring the use of across() to handle the situation where multiple columns will be 'collapsed'. The biggest change would be the need internally to identify columns specified in across() so that any columns not listed could be added to the group_by() call. Changes to single columns as permitted by collapse_col_flex() would be easy to identify as they'd be named arguments in ....

[allenbaron]

Consider allowing user-defined functions to be used as methods. This would probably need to have a well designed error condition to handle cases where user-defined functions return more than one result per row.