I really thought that implementing the watchdog (STWD100NYWY3F) was going to be effortless, but after carefully soldering the components and adding the appropriate code, it just would not work!
Fortunately I had some spare PCBs and watch dog chips so I set up a test rig to diagnose the problem. Using an oscilloscope it was easy to see where the problem lay as the watchdog was actually responding as expected, but the MCU reset line was not responding. According to the STWD100NYWY3F documentation, a 4.7k resistor is required to allow 'signal contention' between the watchdog and the MCU reset and replacing this with a 2k allowed the signal to come low enough to trigger a proper reset:
During implementing code to control the watch dog, I realised that I had to disable the device whilst the BME280 sensor chip was 'doing it's thing', which takes about 3 seconds. The watchdog has a window 1.6 seconds long in which it expects a heart beat pulse from the MCU so, unfortunately, it seems that the STWD100NYWY3F is not the best solution. I'm guessing that the chances of the system crashing during that 3 second long period is very low, but for the sake of selecting the correct component, it's worth making another PCB upgrade. It will also give the opportunity to incorporate a switch in the circuit to isolate the reset so that code can be uploaded without having to remove the MKR WAN 1300 from the board. The new watchdog will be the programmable Texas instruments TPS3430WDRCR for which there is an evaluation board to make life easier.