EDIT: Mea maxima culpa, confusion crept in, my 145 number was supposed to be a much looser guess for actual working full time mathematicians. I miswrote this in the original post as applying to math PhD students, which are much lower. Closer to a 130 median.
ORIGINAL: It's not quite a guess, but I don't have precise data on this exact thing either. Previous studies in this field have consistently found a range of between 140 and 150, and you can probably find those with some Googling if you want to corroborate it yourself. I have a long cached memory of seeing a study where theoretical physics PhD students had an average IQ of 150, which also loosely supports this, since theoretical physics is almost its own form of pure mathematics.
>an individual's IQ is not something set in stone, and can absolutely be improved with training
Most psychological research I've seen says no such thing, unfortunately. Believe me, I would love for that to be the case - one extra point of IQ correlates to roughly $1000 extra income per year in the US, and so if your 20 point claim were really true we could potentially cause a double digit spike in GDP over the next few months just by implementing it in smart ways. But my baseline belief is that study is almost certainly an outlier in a sea of similar studies which support the null hypothesis.