How can I know which is the real maximum voltage in the analog pins so I can do an accurate calibration?.
The most accurate way is to use a precision voltage regulator that feeds only the AREF pin, and then use the external reference mode.
The issue with using the Vcc reference is that it can vary, especially due to switching transients and other loads that are on the power bus. These transients will cause noise in your converted values.
There are some way to force any pin to give me always the maximum votage so I can calibrate every X time the YUN programatically?
I think you might be misunderstanding what I mean by calibration. It does not involve taking the reference voltage and calculating bits per volt, and then scaling that to the pressure reading. Those are all theoretical values. There will always be some part-to-part variations in the reference voltage, the analog converter, the sensor, and any conditioning circuitry. By calculating theoretical numbers, you will get close.
But to really calibrate it, you need to take a series of measurements at a wide range of known pressures. For each known pressure, record that physical value and the corresponding raw analog reading. When you have a bunch of such samples, calculate a linear estimation fit of the points to come up with a slope and offset values (y = ax + b) where "a" is your slope, and "b" is the offset. Once you have those calibration values, you can convert any raw analog reading "x" into a scaled physical value "y".
Using a method like this, you fine tune the theoretical scaling factors, and end up with a more accurate converted value. This will calibrate out any variations on the analog reference voltage, and any part-to-part variations. In reality, the actual reference voltage is not so important (as long as it covers the required range) as it will be taken care of during calibration.
If you are making multiple units, each one will need to be individually calibrated.
So If I take as a reference voltage i.e 4.5V I will get
0.5 V --> -2000 Pa --> 114 bits
2.5 V --> 0 Pa --> 568 bits
4.5 V --> 2000 Pa --> 1023 bits
Am i right?
Theoretically, yes. But you won't necessarily get exactly these values due to the variations inherent in every part of the chain. That's why you calibrate, to get more exact values.
On the other hand with this sensors voltage range I couldn't use the 1.1V Internal reference voltage If I understood.
Correct, the 1.1 volt reference will not be useful in your application, as it will severely limit your range of measured pressures. You could divide the sensor voltage to get it down to the 1.1V range, but that could increase the amount of noise in your samples.