Changed the lume.set(), lume.merge() and lume.filter() functions to take
the argument `retainkeys` instead of `isarray`, and to treat the table
as an array by default (avoiding sparse arrays) instead of treating the
table as an associative array by default (retaining keys).