Client can describe a pose to be derived from an other one, for instance
AIM can be derived from GRIP by applying an offset and rotation.
This reduces the amount of data that needs to be polled and transmitted
over the network.
Add the timestamp where data was requested, use this to compute how far
in the future the application is requesting samples.
With this information, client requests 2 tracking samples: one at the
current time and one in the future.
Server now keeps track of actual samples + one predicted sample, and is
most of the time in the interpolation use case.