Large DAGs/data take a prohibitive amount of memory

[jmafoster1]

Describe the bug
Generating tests for a large DAG/dataset takes a lot of memory. I think this is because a new estimator is defined for each test case, where each estimator has its own copy of the dataset. I think it would be better to have tests as a generator.

To Reproduce
Steps to reproduce the behavior:

1. Take a large causal DAG (50+ nodes)
2. Take a large dataset (10k+ rows)
3. Try to generate causal test cases
4. Watch your memory fill up until the process is killed, or you computer becomes non-responsive.

Expected behavior
The CTF should handle larger scenarios smoothly.