You can run these sorts of tests in a VM with PCIe passthrough.

And hardware companies usually have large test-beds of hardware. So, Nvidia should be able to build a comprehensive test-suite.

the same way they should have been able to build an upstreamable opensource driver all this time, right? lol

No. Cause one (hardware) is old-hat and the other (FOSS) is big and risky and scary.

The reason upstreaming it is considered extremely good really doesn’t have to do with unit testing or really exactly with testing at all. The issue, as stated eloquently by the parent of your post, is that without it being upstreamed, it’s Nvidia’s problem to fix any issues caused by one of their dependencies in the kernel changing, where as if it’s upstreamed, it’s only maybe partially their responsibility (main burden is on whoever changed the dependency).

The kernel has bugs, all software has bugs.

But there are massive massive test suites run by hundreds of companies on linux-next. If you email off a stack of patches that breaks something, chances are you'll get emails back telling you which patch had the breaking change in.