Fill out linalg tests

[kshyatt]

I found a bug in taking the inverse of an Eigen which is hopefully fixed now. I also added tests to get some of base/linalg to 100% coverage.

[andreasnoack]

A long lived one. I made that error in March 2013. We should (again) consider saving the symmetric and general eigensolutions as separate types. It is a shame to invert the vectors when you know that the problem is symmetric, but that is for another pr. Thanks for fixing this.

[kshyatt]

Looks like I got hit by the AppVeyor timeout. Everyone ok with this change?

[ViralBShah]

Nice find. I think given its passing on travis, should be good to merge.

[kshyatt]

But, but, #12086 (comment).

Also I think this PR really demonstrates the value of getting files to 100% coverage, considering base/linalg/eigen.jl had only 5 lines untested and yet this little 🐛 was hiding in there.

[StefanKarpinski]

Figuring out how to get to 100% coverage is definitely something we should do.

[tkelman]

restarted, passing now

agreed with @andreasnoack that symmetric/hermitian eigenproblems should probably be a different type than nonsymmetric some day