fmadd in vec256_base should be on Vec256<T>, not T (#36751)
Summary:
This should have been intended to be the general version of fmadd in
vec256_double and vec256_float.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/36751
Differential Revision: D21148849
Pulled By: pbelevich
fbshipit-source-id: 0805075d81c61d22383a3055aebcb91d09e545de