fix sync of eigenmode calc when no mode is found

[stevengj]

This is an improved version of #594 that skips most of the eigenmode calculation on non-master processes (#595), which fixes a potential deadlock if the processes didn't agree in the case where no mode is found.

Also, it fixes a memory leak when no mode is found (returning NULL). It also gets rid of the rand_r call from #568, which is no longer needed since H is only initialized on the master process.

[coveralls]

Coverage increased (+0.4%) to 82.408% when pulling 013f21a on eigmaster into 97396af on master.

[stevengj]

@ChristopherHogan, it seems like the MPI Python test is hanging on Travis. What do I have to do to reproduce this locally? make -C python check doesn't hang for me, but I'm not sure whether it's using multiple processes.

[ChristopherHogan]

make check uses 2 processes when Meep is built --with-mpi. It looks like binary_grating.py is the hanging test, so you could try running it manually with mpirun -n 2 python path/to/python/tests/binary_grating.py.

[stevengj]

Looks like it is fixed now. I'll merge once the latest CI tests are green.