It is hard to judge what the rocket trying to do (to impute behavior to machine is may be to early). Anyway, the general impression of majority of people commenting these videos is that flight computer is confused.
The cold gas thrusters are used only for roll control at atmospheric decent phase so they should works in pairs. Cold gas thrusters (RCS in your question)are too week even to try to correct excess of angular velocity that killed the booster.
When gambled engine solely is used for attitude control the control forces are applied in single point. This way, the pitch/yawing control moments are coupled with uncompensated lateral force. Consequently, any attempt to compensate significant moments introduced by cross winds or wind shear inserts lateral force and deviation from target path. Imposed sideway velocities induce back aerodynamic pitch/yow moments. On the other hand, any divert move performed by single-ended attitude control is possible only by tricky tilt and set upright maneuver.
Evidently, on certain point, that highly nonlinear feedback becomes pathological and triggered a Stability Loss and hardover beyond recoverability.
Such deadly vacillation is caused of close frequency response of the lateral feedback interfere with the pitch/yaw channel. Figuratively speaking, the guidance computer is in vacillation between contradictive priorities: attitude stability and horizontal deviation.
The thrust of cold gas thrusters may be estimated by their openings but even without calculations one may decide is it possible such two inches nitrogen nozzles may turn back 20 ton, 42 meter long vehicle in seconds.
The thrusters was activated most likely by axial twist after violent contact of legs with platform.