From the depth data reference:
With the unit in landscape orientation, screen facing the user: +Z points in the direction of the camera's optical axis, and is measured perpendicular to the plane of the camera. +X points toward the user's right, and +Y points toward the bottom of the screen. The origin is the focal centre of the color camera. The output is in units of metres.
Notice the unit of the values is apparently in metres. Looking at the actual values observed from the Tango Device it is clear the Z value (depth) is in meters.
X and Y values, however, do not appear to be in meters. This spreadsheet and another of x,y,z values shows actual results from the Tango point cloud. The min/max values of all the point clouds I've seen seem to indicate the units of x,y are in radians since the values are less than 2, but can be greater than 1.
My goal is to convert the observed points to real world coordinates matching the pose data (in Java).
What are the units of x,y?
All the coordinate elements of a depth cloud point - X and Y as well as Z - are in meters. Naturally any triplet (X, Y, Z) must lie within the field of view of the depth camera. The allowable range of X and Y thus depends on the value of Z: X must satisfy abs(arctan(X/Z)) <= fovX/2 (and similarly for Y).
The X and Y ranges you are observing in your sample frames are not angle radians (or normalized in any way); they only appear to be in in that range because of the typical Z distance of your scene. If Z happens to be larger in a different scene then X and Y can also be larger and satisfy the field of view relation.