Add some features that TreeCorr uses.

[rmjarvis]

I have had a similar set of classes in TreeCorr for various coordinate calculations I do there. I'd like to transition over to using Coord and remove the TreeCorr versions of these, but there are a few things I need that aren't in the current Coord package:

• xyz_to_radec converts from 3D x,y,z coordinates to ra, dec in radians. Specifically, this needs to be able to handle numpy arrays. Other than that, it is equivalent to

coord = CelestialCoord.from_xyz(x, y, z)
ra, dec = coord.ra.rad, coord.dec.rad

• radec_to_xyz converts from ra, dec coordinates in radians to x,y,z. Again, accepting numpy arrays. It is otherwise equivalent to

x,y,z = CelestialCoord(ra * radians, dec * radians).get_xyz()

• AngleUnit.valid_names lists the names that are valid inputs to AngleUnit.from_name.
• Along the way I fixed @erykoff 's bug report ResourceWarning in LSST stack #8.

I don't know if anyone will want to code review this. But I'll plan to merge this Friday unless someone speaks up.

[jmeyers314]

LGTM. I listed a few small ideas for you to look at.

[jmeyers314]

Maybe a good opportunity to make a vectorized sincos?

[jmeyers314]

This line's about 10x faster for me as arctan(z/np.sqrt(x**2+y**2)), which I think is okay. (We already know the right quadrants.)

[jmeyers314]

It's about 10% more expensive here to also calculate r=np.sqrt(x**2+y**2+z**2). Maybe make that at least an optional output for symmetry?

[rmjarvis]

Thanks Josh. I didn't make a vectorized sincos. But I took your other two suggestions.

Some grist for the sincos mill:
https://stackoverflow.com/questions/32397347/is-there-a-fast-way-to-return-sin-and-cos-of-the-same-value-in-python
numpy/numpy#2626

I think the upshot is that to do this, we'd have to roll our own in C++ that can call MKL's vdSinCos if it's available. That's more work than I want to do right now. (Mostly in setup.py to find the MKL library if it's installed and have a backup plan if not.)