Execution fails when tries to set no memory limit for model building

[rabsr]

Describe the bug

Execution errors out when None is used for memory_limit

automl = autosklearn.classification.AutoSklearnClassifier(
    time_left_for_this_task=120,
    per_run_time_limit=30,
    tmp_folder='/tmp/autosklearn_classification_example_tmp',
    output_folder='/tmp/autosklearn_classification_example_out',
    memory_limit=None,
)
automl.fit(X_train, y_train, dataset_name='breast_cancer')

Actual behavior, stacktrace or logfile

Traceback (most recent call last):
  File "example_classification.py", line 35, in <module>
    automl.fit(X_train, y_train, dataset_name='breast_cancer')
  File "opensource/auto-sklearn/autosklearn/estimators.py", line 598, in fit
    dataset_name=dataset_name,
  File "opensource/auto-sklearn/autosklearn/estimators.py", line 357, in fit
    self.automl_.fit(load_models=self.load_models, **kwargs)
  File "opensource/auto-sklearn/autosklearn/automl.py", line 1422, in fit
    is_classification=True,
  File "opensource/auto-sklearn/autosklearn/automl.py", line 487, in fit
    task=self._task,
  File "opensource/auto-sklearn/autosklearn/automl.py", line 830, in subsample_if_too_large
    if memory_limit <= megabytes * 10:
TypeError: '<=' not supported between instances of 'NoneType' and 'float'

Environment and installation:

Please give details about your installation:

• Is your installation in a virtual environment or conda environment? virtaul environment
• Python version: 3.7.3
• Auto-sklearn version: development branch

[mfeurer]

Fixed via #1118.