Current sampling 2: from ADC count to amps

The board build (episode 6's bench, with the ADC registers as bench variables) runs the real bsp_get_phase_current() and CurrentCalibrationStep(). At 1197.4 ms U's JDR1 holds 32990: one count is 1.2589 mA (2 mΩ shunt, op-amp gain 20, 16 bits over 3.3 V), minus the zero 32791 gives 199 counts, 0.2505 A; below the zero, an (int32_t) cast makes it negative. The zero is the average of 1024 readings at 0 V on the first enable. Then four real runs: a zero at 36100 fails on the first reading; one glitch voids the whole calibration, yet the same glitch 50 µs later is never seen; after a failure, clearing the error and enabling again recalibrates but never switches the PWM back on (a servohub bug); a joint pushed with 0.05 N·m while it calibrates gets a wrong zero, so holding still the driver reads 0 A while 0.65 A flows in the motor.