Text this: Maximum likelihood estimation versus least-squares estimation in semi-continuous hidden Markov modelling