arbitrary frequency grid for DFT functions in C interface

[oskooi]

Progress towards #1070 (still needs Python interface). Closes #1141.

Adds support for arbitrary frequency grids for DFT functions. There are now two interfaces to all DFT functions: (1) the original equally-spaced grid specified by (freq_min, freq_max,Nfreq) and (2) the new unequally-spaced grid specified by std::vector<double>.

ToDo: Python interface and tutorial example to be added in a separate PR.

[oskooi]

SWIG typemaps for the freq array are added. However, 2 of the 40 tests in the Python suite are failing (dft_energy.py and geom.py). Not sure why calling sim.add_energy in dft_energy.py is causing a segmentation fault.

[stevengj]

As I commented when we discussed #1141, we should probably be using std::vector<double> freqs for every argument where you are passing double *freqs, size_t Nfreqs.

Then if you do flux_object.freqs in Python it will return an array of frequencies (I believe it generates numpy array typemaps automatically).

[stevengj]

The dft_ldos accessor functions were added by @ChristopherHogan in #581. It's a bit weird that we have these for dft_ldos but not for flux objects etcetera? Presumably this code will simplify a lot (and maybe eliminate the SWIG 4 incompatibility #965) if we just have a std::vector object for the frequency array.

Along the way, we probably want to change the dft_ldos object to use freqs rather than omega to be consistent with the rest of the code.

[stevengj]

Note that for each function there should be two interfaces: the old interface that takes (freq_min, freq_max, Nfreqs) and the new interface that takes freqs. However, the implementation of the old interface should just be a one-liner that calls the new interface by constructing an array from (freq_min, freq_max, Nfreqs) . You'll probably want to add a helper function meep::linspace(freq_min, freq_max, Nfreqs) that creates the freqs.

[oskooi]

double *freq is changed to std::vector<double> freq and all C tests are passing. However, it seems that SWIG is not automatically generating typemaps for std::vector<double> freq in Python since flux.freq.tolist() is not being recognized as indicated by many of the Python tests which use get_flux_freqs are failing with:

AttributeError: 'DoubleVector' object has no attribute 'tolist'

[oskooi]

After removing the .tolist suffix to freq as well as some other minor fixes, all tests are now passing. I also fixed a warning bug in python/tests/geom.py caused by python/simulation.py:725. If everything looks good, the final thing to add is a C test.