[Fusion] Fix bug with shared parent selection

[tobias-tengler]

Given the following query:

{
  something { # Subraph 1
    something  # Subraph 1
  }
  viewer {  # Subraph 1 & 2
    exclusiveSubgraphB  # Subraph 2
  }
}

The weighting algorithm will determine it should first plan for Subgraph 1.
The resulting execution step for Subgraph 1 will currently end up with the following selection set as it includes viewer but then doesn't have anything it can resolve from Subgraph 1:

{
  something {
    something
  }
  viewer {
   # empty
  }
}

This PR adds some checks to remove the empty viewer selection from the Subgraph 1 execution step, so it doesn't crash further down the line when attempting to create a request document for the subgraph call.

[tobias-tengler]

Test failures seem unrelated.