Likely:
 {9} Construct more comprehensive test vectors.
 {8} Improve the documentation.
 {8} Add +N addressing (e.g., /foo/,+4d; /foo/+3,/bar/+2 s/foo/bar/).
 {8} Generalize ~ addressing (e.g., /foo/,~4d; /foo/+3,/bar/~2 s/foo/bar/).
 {7} Make lib/regex.c handle [[=x=]] and [[.x.]].  (Alternatively:
       make rx library perform reasonably.)
 {7} Look into improving regex performance.
 {6} Allow for NULs in command script files.

Maybe:
 {6} Add i18n support.  [Need others to do translations.]
 {6} Warn/error for \` and \' in impossible contexts.
       [Only if can cleanly patch regex.c without breaking other apps
        which really want the distinction between these and ^ and $;
otherwise this becomes a {1}.]
 {5} Use tcov to construct more extensive test vectors.
 {5} Allow compilation with different regexp libraries.
 {4} Add -N addressing (e.g., /foo/-3,/bar/+2 s/foo/bar/).

Unlikely:
 {1} Add filters (Use perl instead.)