How to simultaneously fit linear and quadratic lm fit to different parts of the data?

Question

I have dependent variable x, indpendent variable y and group membership idx. There is

1. a linear relationship between x