Label Encoder appears twice for AutoML-generated stacked ensemble pipeline

Currently, for our stacked ensemble pipelines, we are adding a label encoder at the very beginning of all stacked ensemble pipelines for every single classification problem. While this ensures that we have a label encoder for each stacked ensemble pipeline, it also results in a duplicate label encoder if the input pipeline already has its own label encoder. Because these input pipelines are all AutoML generated, they will all have a label encoder in the beginning. An example of this is below:


We have two possible paths of fixing this:
- Remove the lines of code that adds a new Label Encoder to every single pipeline
- Pros:
- If (somehow) we have custom label encoders in the future and each input pipeline has it's own specific version, this will allow that to exist with fewer modifications
- Cons:
- No single source of truth for how label encoding works across pipeline. This opens up the (admittedly very small) possibility that each input pipeline has a different encoded target y as an output.
- Adds additional clutter to the output for what is likely the same component in >95% of cases.
- Remove the label encoders from a pipeline and consolidate with a single new label encoder
- Pros:
- Single source of truth for the label encoder
- Having one consolidated component looks cleaner in the graph
- Cons:
- Does not allow for different types of LabelEncoders in the future to exist in the same grap