Yes, I read the description but these were crashing with zero load beforehand so it's already an improvement. When I tried iperf-based tests before it made zero difference to the incidence of the reboots so I am sceptical but I will take your advice.
FWIW I have two HP 1920-8G-PoE here, one with a recent snapshot including https://github.com/openwrt/openwrt/pull/19365 and the other with https://github.com/openwrt/openwrt/pull/19468
The device with the main branch is still rebooting but the new PR seems to have solved the issue from what I can see:
I'd like to test more but haven't had time to add the timer changes to increase the likelihood of errors. Even so, the results seem positive.