Tighten StridedVector -> Vector in a few spots. Fix #13737.

[kshyatt]

The tau and select vectors in the LAPACK API don't come with
a stride option. I've tightened our requirement to straight
Vector as discussed at JuliaLang/LinearAlgebra.jl#272.

[andreasnoack]

I think that this is fine and nobody will get hurt but I think another solution could be just to check that the stride is one.

[kshyatt]

@andreasnoack it seems dishonest to me to say that StridedVector is an allowed input if only one stride is actually acceptable.

[andreasnoack]

There can be all kind of restriction to the strides (which there already are in the BLAS/LAPACK wrappers) and if someone would like to use part of an existing allocation for one of these vector then a SubArray would be their only option.

[JaredCrean2]

It seems like this is a more general problem of what the right abstraction around a contiguous array view is. When dealing with C code, having vector defined as typealias Vector Array{T, 1} is definitely too tight, and StridedArray is too loose. Maybe this is crazy, inserting a new ContiguousArray in the type hierarchy above Array would solve this problem.

[kshyatt]

Bumping this.

[JaredCrean2]

The more I think about this issue, the more I am convinced solve it requires some kind of interface specification that says "this array can be passed to C". Whether that is an abstract type or a trait-ish thing I'm not sure.

In either case, the important behavior is that pointer(A) returns a pointer that can be used to access the entire array A.

[andreasnoack]

Looks good to me. We might handle this differently in the future but this is definitely an improvement.

@JaredCrean2 See #15308.