Misc linalg tests

[kshyatt]

We had two expm(x::Number) methods defined.

I moved some Hessenberg tests out of dense.jl to qr.jl which is a more natural place for them.

I added tests for cholesky.jl and matmul.jl.

[kshyatt]

We have an unrelated OOM failure and something I don't understand on OSX. Weirdly enough, I ran this locally on OSX and it's fine?

I can reproduce on julia.mit.edu. When I remove the set srand seed things seem fine.

[kshyatt]

In the process of hunting I realized we were running the tests on the same matrices multiple times for no reason.

[tkelman]

And running it multiple times was causing the failure?

[kshyatt]

I honestly have no idea. When I switched the random seeds, it went away. When I didn't preset a seed, it went away. I think this situation the linalg tests are in right now where we run with a specific set of "random" matrices each time is kind of brittle.

[kshyatt]

With the CI passing and the comment to explain how to reproduce, does this look ok, @tkelman & @andreasnoack?

[andreasnoack]

Looks good to me. I''m puzzled by the the NaNs, but I don't feel like spending a day or two on tracking down an issue that has only been seen in a test for a specific seed.