Since the LogMessageStreamer doesn't have an input, I created a custom transformer with the LogMessageStreamer an emailer and just an input (which isn't hooked to anything). I hooked the last transformer in my workspace (a FeatureWriter) to the input and that way the custom transformer doesn't fire off until after the feature writer is done. That way, at least in initial testing, I get to scan all the logs before I email if I find an error.
This seems a bit kludgy so it would be good to know if there's a better way. The errors I'm particularly interested in are logged by the SDE writer which has a parameter to validate features before writing them. I want to log if one fails but move on and not fail the workspace for the rest of the features.