Support for saturating doubling multiply add

There is a saturating doubling multiply add instructions in NEON and SVE - `vqdmla`, `svqdmla` which are equivalent to `hn::SaturatedAdd(hn::Mul(2, hn::Mul(a_real, b_imag)),hn::Mul(2, hn::Mul(a_imag, b_real)));`. In testing, we have seen a significant performance decrease using highway vs NEON and highway vs SVE. I was wondering if it would be possible to add a similar instruction to highway so we are able to leverage these `qdmla` instructions.