Text this: A Closer Look at Invalid Action Masking in Policy Gradient Algorithms