I must say, I’m disappointed too. I really wonder what’s up. If the motor works well after you find the zero-angle using your “extended” method, then it means the problem occurs at init-time, and not afterwards, so it is unlikely to be a slipping magnet or other physical problem.
And the fact you have the same problem with different MCUs and different motors, means it is a software, driver or sensor problem, but not the motor.
One thing I note is that the voltage_sensor_align and voltage_limit are quite low compared to the voltage_power_supply. Why run it at 31V if you’re only using 8V? You will have a better fidelity on the PWM output if you bring these values closer together… But somehow I doubt that’s the source of the problem.