Hello SterlingMinsha , thank you in advance. Meanwhile, I tried another, better damper with 3 Hz resonance frequency (instead of previous 11 Hz).
Also, I found Cross Axis Sensetivity being about 2% instead of 0.2% in datasheet, onto Y axis so Gyroscope Data Post Processing described in part 4.6 of datasheet would not help, as I understand. Also, I found a too big influence of a static tilt around X and Z axis onto Y axis gyro readings, about ten times more than I estimate in thery. All that I tried to compensate in software, adjusting coefficients experimentally.
Now results are substantially better, but I still have about 0.5 deg error (random) of angular position (angular speed integrated) after a couple of seconds of a pretty slow movement and that is still too much for me.
Also I discovered errors are about 2 times less when I suspend the experimental bench on strings instead of moving on the table. In the last case there is no vibrations from friction and movements are smoother.
I would appretiate very much any data what precision of angular position and speed was achieved in reality, for slow movements. My goal is to achieve about 5 cm/sec velosity error after couple of seconds of movement at about 10 cm/sec and 10 deg/sec, little vibrations. And I got doubt if it is possible for BMI270.