Enhance move_first and move_next semantics

[timothy-shields]

The Enumerator<T> concept currently prescribes the following usage pattern:

1. zero or one move_first
   • if zero move_first, done
   • else if one move_first and false, done
   • else one move_first and true, proceed to (2)
2. zero or more current
3. zero or one move_next
   • if zero move_next, done
   • if one move_next and false, done
   • else one move_next and true, proceed to (2)

The Enumerator<T> concept should prescribe the following usage pattern:

1. zero or one move_first
   • if zero move_first, done
   • else if one move_first and false, zero or more move_next (all returning false) then done
   • else one move_first and true, proceed to (2)
2. zero or more current
3. zero or one move_next
   • if zero move_next, done
   • if one move_next and false, zero or more move_next (all returning false) then done
   • else one move_next and true, proceed to (2)

This differs from the previous semantics, which forbid calling move_next any more after having gotten a false return from move_first or move_next. That previous restriction imposed by the semantics was both unnecessary and (it turns out) inconvenient. Operations like merge, union, intersect, except, etc. can be implemented with less state if this restriction is relaxed.