[ ] Save speech to text as new event type. This data is currently being saved as an audio message and it's messing up all kinds of things. We've addressed most of them but the final straw is that it's necessitating loading the timeline in get_chat_list_item_subtitle rather than using the lastEvent data.
[ ] Create a function in pangea_message_event (or modify existing representations getter to allow specifying this) to just get embedded representations. Use this instead and just use message.body if needed representation not present.
This has two fairly big tasks: