Clarify flushing/"blocking" behavior

[Florob]

The SpiDevice trait's documentation is now very explicit that operations may return before they are finished. That brings up some questions:

1. What happens with adjacent operations? E.g. if there are two back-to-back write() calls, is it legal for the second call to return Err(Busy), or is it expected that this call will block until it can enqueue the operation?
2. In the same vein, is calling write(); write(); acceptable for drivers, or does it have to be write(); flush(); write() to not risk an error?
3. What about other traits like I2C? They do not mention that returning early is acceptable. Does that imply that they block until the bus is idle? If not, why don't they have a flush() operation. If so, shouldn't that be explicitly documented?

[eldruin]

Points 1. and 2. have been fixed in #456.
We may need documentation improvements for the I2C case.