Thanks.
My unit does feature the AFCI mentioned by Yaakov on post #18 :



And agree, the arc is not a product of inductance.

...the extreme damage shown in the photo is due to a very poor installation. No PV array should ever be installed in a position where a catastrophic failure can ignite roof material.

Well, that is not extreme damage... This is more like it, carrying total loss of dwelling :

And reason for this thread to explore and learn prevention methods in and out of the box...

The percentage of U.S. dwellings using shingle roofs is very high :
----> https://www.ecosia.org/images?q=solar panel installed

 

Another idea I had but never tested was to have a carbon pile, like a battery tester fully turned on (tightened) in parallel with the switch. When you open the switch, the carbon pile would conduct most if not all of the current. Loosening the carbon pile would raise resistance until current is negligible. Then maybe take the carbon pile offline too.

This one is 1000 Amps Amazon.com: FJC (45118) 1000 Amp Carbon Pile Battery Tester : Automotive

500 to 1000 amps here, starting at $80.00 Amazon.com: Clore Automotive Solar 1874 500 Amp Carbon Pile Battery Load Tester : Everything Else

 

Given that the stated issue is arcing at unintentional openings of the circuit, it is clear to me that what is required is a sensor that can detect the arc disruption signature in the loop current, and then rapidly open the loop at a different point that is suitably protected. That device could be a remote trip circuit breaker, or some other arrangement. I am not aware of exactly what an arc's current disruption signature looks like but I am guessing that it has some high frequency components. Of course, that would be for AC arcs.
If the solar cell loop is isolated from the physical framework "ground" the possibly a change in the potential difference would serve as a valid trigger. And possibly the folks at Seimens , or whatever company makes the devices Yaakov mentioned in post #18, can provide the solution. Certainly any variation not consistent with the inverter load represents a fault of some kind.