cis (and Float64 -> ComplexF64) support?

Some vectorized math libraries (i. e. Intel VML) have a specialized `cis` routine, although I notice SLEEF doesn't.

Is there any possible benefit to adding `cis` support? In the "Future Work" section of the docs, plans are mentioned to support heterogeneous types in the loop and non-primitive types better. Would that be a prerequisite for including a fast `cis` implementation?

I also noted that if I wrote out the loop, things would at least run, but using the convenience functions gave me an error:
```
using LoopVectorization

x = rand(1000)
vmap(cis, x)
```
throws
```
ERROR: KeyError: key Complex{Float64} not found
```

Start by reproducing the reported vmap(cis, x) example and its Complex{Float64} KeyError. Read the documentation's Future Work section and trace how convenience functions handle heterogeneous or non-primitive types. Done would require a decided approach for cis support and confirmation that the example works with appropriate coverage.

Vom Indexierungsmodell aus dem Issue-Text verfasst.