I read Microchip's AN1207 to familiarize myself with SMPSes. So I was wrong, it's a half-bridge design. I think the root cause of the failure was the input capacitors, after removing them and "testing" them quickly with a simple resistance check, one of them has the familiar RC charging behavior all the way to megohms, the other one stays stuck in the 100s of kohms, I guess it started leaking more at its working voltage, blew some input diodes which shorted the AC straight across the whole input side taking the two transistors with it, and then blowing the fuse. It looks like lightning struck inside the fuse.
I also noticed they left one opamp of the LM358 (on the temp sensing circuit) unconnected, I will connect it as a unity gain buffer to 1/2vcc or something like that. Bad design practice to leave stuff floating, it would have cost nothing more to add the connection. I mean it's ugly, they really left the pins floating around the PCB, there isn't even copper! Just holes in the PCB.
And no, I can't replace it with a PC power supply, it's a fridge! Where would the temperature regulation come from?
And also no, MJE13005D aren't that easy to find. None at DK or Mouser or eBay. Certainly none at the (few) local electronics stores left here. The only reason I had 2SC3039s in my parts is because many years ago I repaired a monitor that used them.
Anyways, I ordered new caps, diodes and transistors (PHD13005), and will power up the new parts in the board with my Variac.
I fully expect the thing to blow up again in another 2-3 years. See you then?