Previously, a failed I2C transaction would leave the data buffer
intact, and if one didn't look at the return status they might assume
valid data were returned. This modification should make it easier to
identify failed transactions.

Especially functions that *set* TMR'ed variables *must* use it,
otherwise they only set one of the three copies, which makes the data
invalid.

When a bus error happened, the I2C peripheral's error bit remained
set. This caused it to respond erroneously in future transactions.

Now, when a bus error occurs, we're clearing the error bit and setting
the error flags in the STATUS_BYTE and STATUS_CML PMBus registers.
After the user reads those registers (using commands 0x78 and 0x7E),
their error flags are cleared.