Fix automl to treat binary as categorical when missing values present

[tgaddair]

This is a temporary workaround to handle the following error:

...
 File "/home/ray/anaconda3/lib/python3.7/site-packages/ludwig/api.py", line 428, in train
    **kwargs,
  File "/home/ray/anaconda3/lib/python3.7/site-packages/ludwig/api.py", line 1317, in preprocess
    random_seed=random_seed
  File "/home/ray/anaconda3/lib/python3.7/site-packages/ludwig/data/preprocessing.py", line 1496, in preprocess_for_training
    random_seed=random_seed
  File "/home/ray/anaconda3/lib/python3.7/site-packages/ludwig/data/preprocessing.py", line 197, in preprocess_for_training
    random_seed=random_seed
  File "/home/ray/anaconda3/lib/python3.7/site-packages/ludwig/data/preprocessing.py", line 1677, in _preprocess_df_for_training
    backend=backend
  File "/home/ray/anaconda3/lib/python3.7/site-packages/ludwig/data/preprocessing.py", line 1142, in build_dataset
    backend
  File "/home/ray/anaconda3/lib/python3.7/site-packages/ludwig/data/preprocessing.py", line 1255, in build_metadata
    backend
  File "/home/ray/anaconda3/lib/python3.7/site-packages/ludwig/features/binary_feature.py", line 79, in get_feature_meta
    f"Binary feature column {column.name} expects 2 distinct values, "
ValueError: Binary feature column gender expects 2 distinct values, found: ['Female', 'Male', '0']

Essentially, the feature was treated as binary because it had 2 distinct values, but because there were also NaNs, they were coerced into '0', which led to there being 3 distinct values during preprocessing.

Long-term, we should instead coerce NaNs for binary columns into the correct "false" type by default (or replace with mode). If no false type can be determined, we should treat the column as categorical in automl.

[ShreyaR]

lgtm!

[ANarayan]

looks go to me as well!