If an entry in the log is corrupt, it doesn't get replayed. Yes, and that is exactly the solution the article calls for: you have to design your log format

If an entry in the log is corrupt, it doesn't get replayed. Yes, and that is exactly the solution the article calls for: you have to design your log format so you can determine a log entry is corrupt, so you don't replay it, while still replaying the remaining entries. You can't solve that using write-ahead-logging, or it'd be turtles all the way down.