Ok so I'm re-thinking the whole chain again.
The breath sensor is going to be the MPX2010GSX, non-smt, and has two mounting holes.
For the amplification side of things, the power supply will still be from the arduino's 5V - using two 9V batteries to provide a dual rail supply would work but I'm being bullheaded - and to get the dual rail I've opted to use the Si7661, a voltage converter. Basically, put in ground and Vcc, and you get Vcc, and -Vcc. Cool! Didn't know such things existed. I've not factored in voltage regulation, so a note to myself, I may need to get LM7805 and LM7905 to regulate the voltages.
For amplification, I'm going with the INA114AP, a proper instrumentation amplifier. It can go down to plus/minus 2.25V minimum, so if I provide +/- 5V to it, should work fine. The arduino's range is 0 to 5 volts anyways. Interestingly, all it needs is one resistor to set the gain.
In my tests today, I was getting between 0 and about 4.5mV at max pressure. Note that because my setup has a bleed vent, there would be no way I can get the full 15mv. This may not be a good thing, because I now need to amplify 0-4.5mv instead of 15mV to 5V, much more noise, and much higher accuracy needed.
What I could do is use one of those aquarium valves to provide a variable bleed valve instead. A bleed valve is necessary as I do not want moisture to directly hit the sensor - most of the air moves out through the bleed - hence the pressure is lower.
The alternative is to amplify 4.5mV up to 5V, which has G > 1000! Not a good thing, I think.
Anyways, enough of this. Parts have been ordered, we'll see what happens next week.