I've been seeing inconsistent behavior with the startup log streaming and while looking at the implementation, I noticed a few issues:

• There's a delay between fetching initial logs and subscribing to the pubsub for updates, this can result in lost messages
• We did not have an indicator for "end of log" meaning to accurately discern it, we'd have to subscribe to both startup script and workspace updates and get an up-to-date lifecycle state of the workspace
• We were relying on pubsub telling us when "end of log" is reached, but this message was never sent

With this refactor, we are not dependent on state from the pubsub and the agent tries to log an EOF at the end.

Thoughts:

• Currently we don't do anything special with the EOF message, but it could be used to e.g. show exit status as a final log or something similar

Here's an image where the script has completed, but logs are missing (quite a lot of logs):

[test]