1. One device can take out all others including the uplink-feed for the station
yes, because in case of short time it took to turn separate port of was too link and something parts can be damaged. I am not saying that will be damage as there where both cases - device got burned and stopped working or port was powered down and nothing happened.
2. There is no log message that shows there is a "shortage"-event
we are working to solve this. If log messages will appear that will solve the power down when RouterOS boots up on the device. As there are no way to monitor if power is on the port or not. Will add feature request that events are logged. If it is going to be possible, it will be added as it is useful to know.
3. If luck will leave us there is water in the cable or condensing water at a plug... Happy random booting of all any time?
mount and maintain your devices properly. But short detection is turned on only when power is given on the port, so if short is plugged into the port and it is set on AUTO or ON, it will still attempt to detect if power can be turned on, and if not, it will stay in detection phase, if for some reason resistance is within the correct parameters, power will be given on the port and short and overload detection turned on.
With I2C bus the I/O and ADC of the ATtiny461 is more than enough to do a propper per port current measurement and a single POE shutdown per port
yes it is, that is whey when port overload is detected (by measuring load on the port) single port is turned off. since the tolerances allow to turn the port off in time that board will not be damaged. And that is not the case for short-circuit.
I had a look on the circuit and I think that you will not able to solve the issue.
what issue exactly (you named several).