There are two ways to construct a `DocumentSourceExtensionOptimizable`: directly from a source or transform stage (top-level) or indirectly from a stage that desugars into a source or transform stage.

The former has a parse node as we are able to go from `stageDescriptor->parseNode->astNode->logicalStage` but the latter doesn't as we only have access to the astNode (and go from astNode->logicalStage).

The key difference of not having a parseNode during full parse is the inability to get the `queryShape`. This is ok though for the indirect construction case because the top-level desugar stage will be responsibly for `queryShape`.

Because of this, we should have two classes for `DocumentSourceExtensionOptimizable`: one that is "ShapeAware" with a `parseNode` (`DocumentSourceExtensionOptimizableShapeAware`) and another that only has the `logicalStage` (`DocumentSourceExtensionOptimizable`). `DocumentSourceExtensionOptimizableShapeAware` will derive from the base `DocumentSourceExtensionOptimizable`.