Agent execution errors are not propagated to the caller

[asaf]

An agent runner runs on the background, producing events, during an execution failure, an error is not generated as event, causing errors to be silently ignored.

This is related to issue #56 where the introduced fix was to protect specifically from creating a non existing agent but other errors may occur in runtime (such network error, etc.)

Should generate error events and be seen in the browser

[ErezSha]

@asaf
The way to do this would be to encase each agent run in a try/catch block.
I'm not sure we want to do that, as it would be a generic error handling instead of the user from handling the error by himself , like we do in our examples.
I think that error handling the agent's run itself is best kept in the user's implementation, IMO.
There are a couple of examples I think I would want to do as a user:

1. retry mechanism
2. running a different agent if the agent fails

both of these can't be done if we do our own error handling.
The above is not entirely correct, as the user can do their own try/catch in the agent and not raise an error.