[FEA] Add outer() for outer product

[cliffburdick]

Use cuBLAS *ger functions

[luitjens]

If we have something like outer_impl(A, X, Y) where A is batch_dims x N x M and X is batch_dims x N and y is batch_dims x M then we should be able to do this pretty easily by having outer internally clone X to X is batch_dims x N x 1 and clone Y to batch_dims x 1 x M then call matmul_impl.

Note we should probably also support the axis parameter to match matmul/matvec.

[luitjens]

@cliffburdick fixed?