For eig, when eigenvector=False, do not force the row-major-contiguous input to be column-major-contiguous because A and A^T have same evalues.

For svd, when input is row-major-contiguous, calculate svd as if svd is done on the input's transpose, and then return r0^T, r1, r2, where (r0, r1, r2) are from magma gesdd.

cc @ngimel @jianyuh @nikitaved @pearu @mruberry @heitorschueroff @walterddr @IvanYashchuk @VitalyFedyunin