Both open points are addressed; implementing SpiBus was a bit of a backwards exercise, but in the interest of keeping things easy to maintain and efficient for the case where it matters, I think it's a sensible choice to effectively route everything through an implementation that primarily caters for SpiDevice use cases.

The test is more than a build test but less than a functional test (I don't even have the wires with me to loopback MISO and MOSI), but given that the one thing that went wrong earlier was misconfiguring the SPI device in a way that would indefinitely block any transaction, I think that's good enough.

(ad review request: Yeah I know I have pending reviews myself on the UART side; got that branch open again already!)

I think I have addressed all comments. In particular, you may want to have a look at fc0914f.

Many kudos for reviewing this! That helped a lot both in making the API better and gaining confidence in riot-wrappers as a whole.

I think all comments are addressed (one by an elaboration, one by 481c6b3); another commit (cd40adb) fixes build issues.