While adding ENDSTOP_NOISE_FILTER, I try to reduce the impact on accuracy as much as possible by reducing the endstop_poll_count to 2.
Possible fix for #8.
Thanks to @notching, @mensikv and @md66pt for the valuable feedback!
Using TMC2208 in standalone mode, I've inverted the stepper direction to match Anycubics pin layout. M110 memory watcher was removed due to redundancy and - ironically - to save memory.