Restore current multiplex after nested execution

[ydah]

Nested GraphQL execution currently clears the outer execution's fiber-local multiplex.

This can happen when a resolver executes another schema, for example while triggering subscriptions or composing an internal GraphQL API:

before = GraphQL::Current.operation_name
InnerSchema.execute("{ value }")
after = GraphQL::Current.operation_name

before # => "OuterOperation"
after  # => nil

Both GraphQL::Execution::Interpreter and GraphQL::Execution::Runner replace Fiber[:__graphql_current_multiplex] when execution starts, then unconditionally set it to nil in ensure. As a result, the outer multiplex is lost after the nested query finishes.

Besides breaking GraphQL::Current.operation_name, this can cause later dataloader runs to lose their current trace because they read it from the same fiber-local multiplex.

This PR saves the previous multiplex before starting execution and restores it in ensure. Top-level execution continues to restore nil, while nested execution returns to the outer multiplex.

[rmosolgo]

Thanks for this fix!