If we only check the sensor state flag, when a abrupt shutdown occurs,
we'll have to run through the sensor event state machine at least 2
times before achieving a state that causes the P12vGOOD flag to be
cleared.
Since the payload task runs at a greater frequency than the sensors',
it'll try to power-up the payload again based on the erroneous sensor
state

There's a drop in the voltage between the standalone voltage entry and
the value measured at the INA220, which is about 0,6~0,8v, this was
causing the board to always stay between Low-crit and Low-noncrit state.

Reset the power good flags to avoid the state machine to start over
without a new read from the sensors.
This was causing the payload to quickly turn on after a QUIESCE event,
since the INA220 sensor task task a little longer to update all the
sensors readings.
Fixes #23

Now we always jump to PAYLOAD_SWITCHING_OFF state before going to
PAYLOAD_NO_POWER, this way we can safely turn off the payload and warn
the MCH about the failure

Now each sensor must call sdr_insert_entry() upon initialization to
include its sensors in the SDR table.
The SDR table cals realloc each time a new entry is requested, so it
doesn't use unnecessary memory.

INA220 SDR was put in a different source file (ina220_sdr.c) because
it's too large and would polute the original source (ina220.c)

The comparation would work if the SENSOR_STATE_NORMAL wasn't equal to 0,
which invalidates the statement.
Now checks each state a time.

As stated in PCA9547 datasheet:
The fourth bit in the control byte must be set to one in order to enable
the access to the selected channel

Also format the whole file

FMC1 and FMC2 had their channels switched

Fixes #21

Fix for last commit