Avoid discarding already read bytes on error

[jannic]

Closes: #782

It would be preferable to avoid the code duplication between UartPeripheral and Reader. However, before starting a bigger code restructuring, I'd like to get some feedback on the general approach:

• Is it the right trade-off to reserve some memory in UartPeripheral and Reader to store the read error?
• What are alternative approaches?
  • Ignore the issue, accept discarding bytes on error. (Might be acceptable, assuming that serial errors are rare and the received message is probably garbled anyway. Should be clearly documented.)
  • Don't implement embedded_io::Read at all. (Breaking change, lost functionality, IMHO not acceptable)
  • Don't implement embedded_io::Read directly on UartPeripheral and Reader, but instead on a separate wrapper type. (Breaking change, slightly less nice API, zero-cost when not using embedded_io::Read)

[thejpster]

I think this is quite reasonable.

[thejpster]

Leaving this for a second opinion. @9names ?

[ithinuel]

If the UART has/is using a FIFO, I’m afraid there’s no way to tell when did the error occur.
So if we get a parity error for example, we cannot tell which byte is corrupted.

[jannic]

IIRC the error bits are also stored in the FIFO, together with the data words they belong to.

[9names]

[9names]