The "AHRS" orientation estimate that the UM7 calculates isn't great, but the IMU sensor data (i.e., acceleration, gyros, magnetometer) is fine, as long as you've done your calibrations properly. So what you could do is feed another filtering package (such as the madgwick filter) your IMU sensor data and take the orientation output from that. It can be a little bit tricky getting everything set up, so if you have any more questions feel free to post them and I'm happy to try and answer.
Edit (Aug 7th, 2018): In response to @aarontan's comment: I'm not super familiar with the razor, but I will say that a lot of these devices use the same underlying hardware (magnetometer, gyros, accelerometers), so if you plan on running the imu data through a madgwick filter node to get your orientation I doubt it will really matter which you choose. If you plan on using the on-board AHRS for orientation, I would do a search and see what others have said re. the accuracy of the razor. Regarding the UM7 AHRS - in the past it has been notoriously bad, but Redshift labs recently (within the last week) released a firmware upgrade where they report "improved heading estimation, solution for random yaw drift issues, quicker EKF convergence, among other improvements" - I haven't had time to test it yet but it sounds promising.
This is just convention and can be fixed simply by rotating the outputs to correspond with the ROS expectations