Consistency in Rows/Columns/1d-arrays

[odunbar]

It seems like in general when we have tasks requiring linear algebra operations on arrays, it is good practice to ensure that we have a defined dimensionality of the arrays. I would recommend we always use 2D arrays (columns or rows) over 1D vector-type arrays if we are going to use them for linear algebra. Typically these arising from a 'vanishing dimension' of a normally clear 2D array.

E.g when dealing with ensemble based methods, we perhaps wish for arrays to be [ensemble size x parameter dimension] but if we have parameter dimension of 1 then often one defaults to a 1D-array, from which the ambiguity of "which dimension vanished?" arises. This is cured by specifying 2D inputs of [ensemble size x 1] or a [1 x parameter dimension].

The overhead then to pick the dimensions consistently throughout CES to avoid multiple transpositions as we pass through different objects or functions.

note to self, Julia encourages the use of permutedims(A,(2,1)) over A' for applying a transposition otherwise one changes the type into an Adjoint LinearAlgebra object.