Dpocon and Dgecon mismatch

[btracey]

Copied from https://github.com/xianyi/OpenBLAS/issues/664

Our tests at Gonum (github.com/gonum/lapack) for Dpocon compare to the result for Dgecon. For most matrices the two outputs match to machine precision.

The matrix
[2.9995576045549965, -2.0898894566158663, 3.965560740124006,
-2.0898894566158663, 1.9634729526261008, -2.8681002706874104,
3.965560740124006, -2.8681002706874104, 5.502416670471008]

Has a significantly different answer for the two routines. Dgecon gives 0.020928829230190116 while Dpocon gives 0.024054837369015203.

See https://github.com/gonum/lapack/pull/59 for more information