[saimo]

Minor improvements after checking the NXP documentation and the Microchip datasheet:
* removed the START/STOP conditions after the fallback reset (it didn't make sense already, but I had tried to stay faithful to ATMEL's datasheet);
* shortened the delay before generating the START condition (it was too long for a restart and the appropriate delay was already applied by GenerateSTOP()).
Both these changes should not have any relevance to whether the code works or not, as the fallback reset probably is never executed and the other change only improves performance in a negligible way.
It looks my friend might run the test shortly... crossing fingers...