Temporary duplication of capabilities in PDMats

[dmbates]

I'm experimenting with using PDMats to represent covariance matrices for random-effects in the MixedModels package. Because I tend to use the factor more than the positive-definite matrix itself I will experiment with a different representation. When I am finished experimenting I will see how much overlap there is with this implementation and discuss whether to merge capabilities.

[lindahua]

I am now adding another type under AbstractPDMat that allows people to use a factor matrix instead of the covariance itself.