I think the second problem happens when the user is connected (websocket) to one pod, and the message is sent to the API on a different pod. Is that a known issue?
Maybe this feature is not offered on the open sourced version? Scaling mattermost to multiple servers · Issue #9080 · mattermost/mattermost-server · GitHub
UPDATE: confirmed this is a feature for E20 version.
So we are left with the first issue of the bot not receiving all messages.