It is actually an interesting concept.

A hobbyist has little money, so they fight, and fight, and fight with their board.

For commercial operations, probably a significant percentage of chip manufacturers have gotten their chip dropped because the eval board (with the included documentation + code) had too bad quality.

It's a question of risk management. If it takes a week extra to get the evaluation board working, you don't want to find out how much unplanned time it takes to get the own product shipping. What if the datasheets also contains critical errors? The application engineers can't explain why the processor does one thing, while the documentation says it should do something else?

Good samples not only helps as a additional information when reading through all documentation for an UART. They also helps proving that the header file(s) aren't completely broken.

And when evaluating a new processor, it really isn't the UART you want to evaluate. Most probably it is some complex situation you worry about. Figuring out the true speed of the system with some form of simulated load. But you want to be able to directly start modeling this situation, instead of first having to get basic interrupt handlers etc to function.