I reviewed this, whacked it around a little, and pushed it.
Possibly the most controversial thing I did was to move the existing documentation entry for string_to_array() into the string-functions table. I did not like it one bit that the patch was documenting string_to_table() far away from string_to_array(), and on reflection I concluded that you'd picked the right place and the issue here is that string_to_array() was in the wrong place.
Also, I pared the proposed regression tests a great deal, ending up with something that matches the existing tests for string_to_array(). The proposed tests seemed mighty duplicative, and they even contained syntax errors, so I didn't believe that they were carefully considered.
