Make the extracted DAG executable

[stefan-grafberger]

Description

• Change the format of the DAG to something that can be executed directly
• A DagExecutor is required, similar to what I've done previously here
• We can use partial functions and the original implementations in a first version, we do not need to immediately start with DuckDB execution
• Some operations like binops like + and > are not included in the DAG but need to be supported for this

[stefan-grafberger]

Done

• Made updates to the DAG to include comparisons, arithmetic ops, and logical ops

  Next todo:

• Include arg indices on the edges so we know the argument order, see here and here
• Also, think about how to model transformers in the DAG. At the moment, the DAG only includes the dataflow and the model training, but ignores preprocessing with state like OneHotEncoders that are fitted first. There are two main options: we can include these edges in the DAG, too, or we can have a Query State Storage where we have a map from code location/dag node id to fitted transformer. I am not sure yet about what is the best way to do this.

[stefan-grafberger]

Next todo:

• Add function call attribute to DAG node and use it to save lambda functions that can be used to reexecute a pipeline
• See how we can update tests so we can still test whether the DAG looks correct (maybe there is something like Comparator.Any we can use in tests, or we can set the function call to None before comparing the DAG for equality like we do currently)
• Add a new artificial DAG node type that can be used to "extract" (named) pipeline intermediate results to outside the pipeline. Basically, something that can capture the results from operations like aggregate, that we want to store after a pipeline run in some DAG execution result map.
• Finally, create the actual DagExecutor
• Test the DagExecutor

[stefan-grafberger]

Everything but one thing works now as descriped above. The item

• Add a new artificial DAG node type that can be used to "extract" (named) pipeline intermediate results to outside the pipeline. Basically, something that can capture the results from operations like aggregate, that we want to store after a pipeline run in some DAG execution result map.

will be worked on as part of issue #3