dask broadcast variable fails with key error when calculating subset of pandas dataframe

Question

I have a pandas data frame and want to apply a costly operation to each group. Therefore, I want to parallelize this task using dask. The initial data frame should be broadc