Update set_time_index code path for Dask dataframes and impacted test

[rwedge]

Fixes #913

This PR updates Entity.set_time_index so that Dask dataframes do not exit the function early. Specific points in the code where costly validation checks occur have different logic for Dask frames.

This is necessary to fix the test_build_es_from_scratch_and_run_dfs test.
For that test, I also changed the logic to get the two dataframe indexes to match and removed two primitives that have the uses_full_entity attribute. Features using these primitives were returning different values, which I think is happening because the dask entity data is not sorted

[rwedge]

I think we may be able to simplify further by changing this line to

if isinstance(self.df, dd.core.DataFrame) or len(self.df) == 0:

and removing the conditional logic about old_vtype

[thehomebrewnerd]

If you do that, won't you always force a compute() to get the length if self.df is a Dask dataframe?

[rwedge]

I think if the first statement of the or is true, the second is not evaluated

https://docs.python.org/3/reference/expressions.html#boolean-operations

[frances-h]

Not sure if I'm understanding correctly, but we could get into trouble calling .head on a dask dataframe since it by default grabs the first partition which could be empty. Since we're requiring vtypes when creating an entity, using the vtype provided to figure out the index type seems like the safer route.

[rwedge]

I'm proposing the logic look like this

        if isinstance(self.df, dd.core.DataFrame) or len(self.df) == 0:
            time_to_check = vtypes.DEFAULT_DTYPE_VALUES[self[variable_id]._default_pandas_dtype]
        else:
            time_to_check = self.df[variable_id].head(1).iloc[0]

Which, if the dataframe is a dask df, will use the vtype to figure out the index type

[frances-h]

👍 LGTM