Add routine to Fourier transform collection of (x, y, flux) sampled randomly

[psheehan]

Per our discussion over e-mail, this pull request is meant to add functionality to galario so that it can take the Fourier transform of an "unstructured image", i.e. a list of (x, y, flux) where (x,y) do not need to fall on a regularly sampled grid. It does so by triangulating the collection of points and using the triangulation to interpolate onto a regular grid. Once that is done, the regular routines in galario can be used to do the Fourier transform and then to sample onto the relevant baselines.

[psheehan]

On my end, the code seems to work, but needs to be restructured to better fit the organization/layout of the galario code. The changes I think I need to make (though input/feedback welcome):

• [x] Rename many of the new functions (interpolate_to_image, find_triangle, find_triangle_directed_walk, find_triangle_bruteforce) to _h, with the intention of eventually adding _d counterparts for symmetry (if efficient on GPUs)
• [x] Break interpolate_to_image into several sub-functions for ease of reading
• [ ] Currently there is only a sample_unstructured_image, but should there eventually be others like chi2_unstructured_image (for example) for symmetry?
• [ ] Should any of the functions be exposed in the interface for experts? Perhaps interpolate_to_image?

[psheehan]

Ok, I spent some time today renaming most of the new functions to have an _h suffix to indicate that they are run on the host. I also split interpolate_to_image into four smaller functions (triangulate_h, bin_triangles_h, interpolate_on_triangle_h, and interpolate_or_bin_to_image_h), and renamed interpolate_to_image => unstructured_to_grid_h. I think these are more or less the smallest pieces that these functions can reasonably be broken up into... but I could imagine that changing if we ever look into whether is part of the code could go onto a GPU.

Also added a few test scripts that I've been using to test that the code works properly.

A few questions I still have/places I still need help that might require some discussion:

• [ ] Do the new functions all look good? Do any need to be further split up?
• [x] Currently there is only a sample_unstructured_image, but should there eventually be others like chi2_unstructured_image (for example) for symmetry?
• [ ] Should any of the functions be exposed in the interface for experts? Perhaps unstructured_to_grid_h?
• [ ] Is there a better/more formal place/way to include the tests? I see a python/test_galario.py and a src/galario_test.cpp, and also a python/speed_benchmark.py - not sure if it would be ideal to add similar tests to whatever is done in those?
• [ ] Is there anything else the needs to be done to get this merge-ready?