I did this recently, and I'd suggest:

* data wrangling bits are shared
* one blocking top level surface
* one async top level surface

https://github.com/jamesmunns/max31855-rs