test_mpb.py fails with LAPACK 3.10.0

[ahoenselaar]

TestModeSolver.test_compute_field_energy fails because expected_fp and the actual data have opposite signs. The D field data also has opposite sign.
The tests in the MPB repo don't seem to encounter the same problem.

[stevengj]

There is an arbitrary phase choice in the LAPACK eigensolver (which affects MPB), so it's possible that from one LAPACK version to the next we get a different sign out of the eigenmode fields in MPB.

(Though I thought we try to canonicalize the phase to eliminate such arbitrariness?)

[ahoenselaar]

Is the proposed fix then to update the assertions in the test to account for the arbitrary phase choice?