There is an answer explaining in simple terms how a router works translating requests from the local network to outside and back (https://superuser.com/questions/105838/how-
It depends.
According to Section 4.3 of RFC 4787, the UDP timeout of a NAT should not be smaller than 2 minutes (120 seconds), except for selected, well-known ports. In practice, however, routers tend to use smaller timeouts. For example, OpenWRT 14.07 uses a timeout of just 60 seconds.
For TCP, the timeouts can be much larger, since TCP connections are usually terminated by an explicit FIN/FIN-ACK exchange. For established TCP connections, Section 5 of RFC 5382 specifies a timeout of no less than 2 hours 4 minutes (7204 seconds), and OpenWRT uses 7440 seconds.
Concerning your second question, most NATs maintain mappings that are specific to a pair of endpoints (socket addresses). If a host A inside the NAT sends a datagram to socket adress B, then the mapping will only apply to communication between A and B — a different host C outside the NAT will not be able to use that particular mapping to send data to A. (Some so-called full cone NATs allow that, but they are fairly rare.)