Tracking down the exact cause of this particular issue is going to make a new bald spot, so please be patient and diligent by rechecking and only doing one change at a time. It is amazing what little thing can cause such big headaches and even more confusing when you realize that yes you did check it but not when asked about a particular condition. I know it is frustrating and its very easy to dismiss things as already done.
Are you using a fully managed switch? If not then If possible get a cheap cisco 2960S, (ebay for about 150$ usd) these are fully manageable and provide great logs of what happened if configured correctly. These are also 12v dc on the inside so if you dont have AC power available then a simple buck converter (300w minimum) will get you running.
If you have a surge suppressor on that ethernet link, remove it and test. Depending on how you are powering the radio, these things can cause issues. I do not suggest to run indefinitely without surge protection but this eliminates a possible point of issue.
Not sure how you are powering the radios, but we use Packetflux’s Gigabit Sync Injector, supplied with both 24v and 48v (dual voltage) and set the jumpers inside to match the requirements of the radio (we lable each port so we dont fubar a 24v radio on 48v!), when an extra surge suppressor is required (top of tower mounted radios) we use UBNT SP-ETH suppressors. We find these ones to be reliable and since they are very cost effective (about 1/2 the cost of Cambiums and about 1/4 of pretty much every other one), when one goes or is causing problems, its not a big deal to swap it.
There is a chance that it is either the switch or the AP that is damaged (near field ESDs do take out equipment too and you may not ever see or hear the discharge!). For the switch, try moving to another port, at least 4 ports away. Most switches use chips that share controllers on 4 ports, so if in port 1 then move to port 5 or farther. If its the AP, you can set the switch port to only negotiate 10/100 (yes, I hear the groaning) and if the ethernet link still halts then lock the AP to 10/100. This eliminates half of the ethernet issues and provides reasonable service until you can swap/upgrade APs.
Last thing to think about is spanning tree issues. If you have a spanning tree aware switch, any loop can shutdown a port (or just one vlan on a port is PVST is used) and not always the one you think it should be! If you have dual connected clients (two SMs at one site) and you are not using a L3 dual uplink device to combine the two links then the loop will be visible to spanning tree and cause random havoc.