If you look at the difference between PIM Dense Mode and Sparse Mode, that will give you the answer to why the RP is required. In Dense Mode, multicast traffic is flooded from the source across any PIM Dense Mode enabled links. This allows it to eventually reach multicast receivers registered to IGMP queriers. Dense Mode will then prune off PIM links that don't require that multicast group. So, traffic is essentially flooded across the network to the receivers and then unused links are pruned to create a tree from source to receiver.
Sparse Mode uses an RP, which is a predefined router on the network that the source and receivers will built a multicast tree to, and traffic can then be send from source to receivers via the RP. The sources send their traffic into the network and the multicast routers build a path to the RP. When receivers register for a group with their IGMP querier (usually their default gateway), that router also builds a path to the RP. Traffic can now flow from source to receiver without being flooded out of every link. Once this path has been formed the receiver's gateway will request a direct path from the source and eventually forwarding will move over to this path, which may bypass the RP. So the RP serves two purposes, to allow IGMP queriers to discover and build trees from the multicast source, and this allows multicast to be forwarded without flooding the network.
The source IP of a multicast packet is the sender's unicast IP address. The destination IP is the multicast group address.