I think you're giving a bit too much credit to game logic codebases.
They generally move at a pretty fast pace and change often(due to the iterative nature of what's "fun"). I don't think I saw a single unit test until I was well out of the game industry.
My go-to exception for this has always been Factorio. They have not only unit tests, but integration tests, regression tests, fuzzers, performance regression tests, ....
Their blog regularly goes into details about how it all works. It's very much worth reading.
It seems to me if the physics engine and whatnot is deterministic I don't see why it wouldn't be possible to record input sequences and keep those and their effects as regression tests.
Those tests can be written, but if you do it naively coverage is measurable and very small.
And the physics engine is not necessarily deterministic (or if it is, you can't always control all it's parameters)
It's been a few years but if I recall correctly we did have a "smoke test" running on a jenkins box under someone's desk that booted a few levels with most of the game entities(they were called danger_room_1/2/3/etc) and verified that the game didn't crash. Aside from that though there wasn't much else which lead to some spectacular build breaks.
They generally move at a pretty fast pace and change often(due to the iterative nature of what's "fun"). I don't think I saw a single unit test until I was well out of the game industry.