Why Generating No Evidence is Not a Good Solution When Encountering Zero Events: Reducing Bias From Monotone Likelihood in Rare-Outcome Cox Models Using Penalization