Fix Wire library regression plus a little optimization

[MX682X]

also some adjustments in the documentation

[MX682X]

Since you are still working out the details on the core (I guess), there is one more thing I stumbled upon by chance:

DxCore/megaavr/cores/dxcore/WInterrupts.c

Lines 153 to 154 in 94ff095

	"in r0, 0x3b" "\n\t" // RAMPZ
	"push r0" "\n\t" // on the stack.

DxCore/megaavr/cores/dxcore/WInterrupts.c

Lines 220 to 221 in 94ff095

	"pop r0" "\n\t"
	"out 0x3b, r0" "\n\t"

You are always pushing RAMPZ, even though only a handful of parts even have this register. I suggest to use the built-in gcc define__AVR_HAVE_RAMPZ__to save some bytes on the smaller parts by checking if it is defined

[SpenceKonde]

Thanks, as you can see i was thinking only of the AVR128Dx parts when writing that library.

[SpenceKonde]

fixes #396 #401 #393

[SpenceKonde]

Yo we got a little problem here. There are currently differences from megaTinyCore version. I don't know which ones are the correct ones in all cases:
Wire.cpp

 *@return     size_t
 *@retval     amount of bytes copied
 */
size_t TwoWire::write(const uint8_t *data, size_t quantity) {
  twi_buffer_index_t i = 0;
  for (; i < quantity; i++) {
    if (TwoWire::write(*(data++)) == 0)
      break;   // break if buffer full
  }

  return i;
}

vs

 *@return     uint8_t
 *@retval     amount of bytes copied
 */
size_t TwoWire::write(const uint8_t *data, size_t quantity) {
  twi_buffer_index_t i = 0;
  twi_buffer_index_t qty = (quantity >= BUFFER_LENGTH) ? BUFFER_LENGTH : quantity; //Don't overfill the buffer.
  for (; i < qty; i++) {
    if (write(*(data + i)) == 0) break;   // break if buffer full
  }

  return i;
}

Not sure which direction this should be merged.

twi.c

/*@retval     error code, if enabled. Not 0 means something went wrong
 */

vs

/@retval     amount of bytes that were written. If 0, no write took place, either due
 *            to an error or because of an empty txBuffer

Which is correct? this sounds really familiar.

vs

I'm planning to add dirty_tricks.h to DxCore, and was thinking to make Wire more portable by putting

#ifndef _fastPtr_y 
(definition copied from dirty_tricks.h)
#endif             

Does that sound reasonable?

I'm guessing we want the DxC version of this one, and assume it's what fixed the bug?

    _data->_bools._toggleStreamFn = 0x00;

vs

    _data->_bools._toggleStreamFn = 0x01
;

[MX682X]

about write (buf, length) - The changes that are part of this PR make the array write about 22 bytes shorter. Before that, there was an icall to write (byte). And as we check for bufferoverflow in the write(byte), there was no need for a second check.

the return value documentation was not corrected since the publishing, so it was still about the incorrect implementation. So I decided to fix it while going through the code. The Wire.cpp file is documented correctly.

and _toggleStreamFn = 0 is the fix, as explained in the discussion