Advanced Agent memory

[julian-risch]

We can build on the Agent implementation in haystack-experimental, in particular it's state.
Currently there is no serde for state (removed in this commit)

[sjrl]

Just to add my thoughts on this. I think State is part of the memory conversation, but I think we should also talk about/consider how to store the ChatMessages returned by an Agent after it exits the loop. I think these messages are also relevant for memory in multi-turn conversations and for restarting an agent if we were to add break points.