For such low RPM there is not enough EMF generated for reliable and accurate feedback, and simple HALL sensors for the phases are also not good enough anymore.
So there are 2 options left.
First option is to add a high resolution encoder. This can for example be a small magnet with a HALL sensor that measures the direction of that external magnet (AS5600 is an example of such an encoder) Higher resolution optical GRAY encoders also work.
The other option is to use no feedback at all, but usemore current then you need to force a magnetic field and hope the rotor follows it. This effectively turns your BLDC motor in a 3-phase stepper motor.
This second way does not need extra hardware. If you have simple phase HALL sensors, then you can use these to make a phase difference guesstimation. In this second case you can not go to 90 degree phase difference for FOC, but you need to go to a lower angle, for example 45 degree. This of course lowers efficiency of the motor, but it the motor encounters some unexpected resistance, then that resistance increases the phase angle (to for example 60 degree) and at that angle the motor delivers more torque, so the rotor keeps folowing the magnet field of the stator.
If you use full FOC with 90 degre phase angle, then when extra torque increases the angle over 90 degree, the torque that the motor delivers decreases and you loose syncrhonisation between rotor and stator.