We have some time_us variables laying around, for example in the blocking i2c implementation

I think it might be better to use the MicroSecond/MilliSecond structs defined in time.rs instead, for reducing the risk of errors, and making clear what the unit is at the call site