Rewrite solve and dot with diagonal matrices

### Description

We have special rewrites for Det and MatrixInverse, but not for Solve/Dot. Simplifying these cases should allow bridging the gap between things like batched univariate normal and multivariate normal with diagonal covariance