This is somewhat technical, but I’ve been using a special tool from Google that allows one to test bugs against arbitrary versions of Chrome, using a bisect algorithm to efficiently isolate the problem:
https://www.chromium.org/developers/bisect-builds-py/
I’ve tracked down the introduction of this issue to this change:
https://chromium.googlesource.com/chromium/src/+/4a76c96d2a4021257e691058c1fe1fb60d9d8327
It is the implementation of this proposal to change the way that block elements are handled inside inline elements:
Stunningly, the document makes it clear that they knew that it was a breaking change, but that they went ahead with it.
I will shortly run the bisect process again to find where the problem was fixed in v102.