should close #127209

Because the FORK output is a union of the subplans' outputs and we keep the same Attribute objects, we don't always apply the right optimizations.

Take this query for example:

FROM search-movies METADATA _id
| FORK ( EVAL a = 1 | WHERE _id == "1" )
             ( EVAL a = 2 | WHERE _id == "2" )
| KEEP _id, _fork, a
| WHERE a == 1

When we apply constant folding, we incorrectly determine that the a attribute from the last WHERE condition is a constant. In cases like these we either end up not returning any rows or we return the output of both FORK branches.

To fix this, the FORK output will no longer reuse the same Attribute objects from the fork subplans when we do the union of output attributes.
Instead FORK will output a new set of attributes that have the same name and type but with different IDs (in a way similar to STATS).