Out of the box, DoubleFloats work with these LinearAlgebra functions:

```
isdiag, ishermitian, isposdef, issymmetric, istril, istriu
norm, det, dot, tr, condskeel, logdet, logabsdet
transpose, adjoint, tril, triu
diag, diagind
factorize, lu, lufact, qr, qrfact
```

I want to support `eigvals`

, `svdvals`

. I can obtain Float64 approximations by converting the given matrix to Float64 elements and running `eigvals`

or `svdvals`

on that and then converting the result to DoubleFloat elements.

(a) is there an efficient way to refine the approximations to more accurate results?

(b) is there an efficient way to use the functions that work already to get `eigvals`

, `svdvals`

?