Hi,
First, is your application indoor or outdoor? Be aware that the Xtion doesn't work outdoor, otherwise you may want to look for a stereo camera.
1) Your first approach would give you odometry poses, as the second will correct the odometry poses on loop closure detection. You got:
sensors ----> odometry (/odom to /base_link) ----> rtabmap ----> updated TF /map to /odom
2) You can use TF for convenience to get the current pose in /map frame (/map -> /odom -> /base_link). What you can do is to subscribe to your odometry output to know when a pose is available (in /odom frame), then use TF to get it in /map frame.
cheers