Tinygp

[sblunt]

Continuation of #359. I gutted the radvel GP API to use Dan Foreman-Mackey's new tinygp package, which is super flexible in terms of defining custom kernels. This is a pretty big change, so I'd consider releasing a new version number.

A few notes:

• the new API I wrote currently doesn't support celerite, so if we want to add that back we'll have to reimplement it
• the GP code is a little slow right now, I think because I'm doing a lot of slow indexing. I wanted to PR this before trying to optimize the code, but it should be profiled and improved. I don't think it will be too much work to speed up.
• Some of the notebooks don’t appear to be working but I don’t think that’s my fault (the GP one works).
• Also the ic_compare n planet test in the unit tests isn’t working on my computer— I don’t usually use that functionality, so not sure what might be going on there.
• thanks to Dan Foreman-Mackey for helping me get up to speed with tinygp!

[sblunt]

Oh one other thing-- I found and fixed a bug with the gamma subtraction that looked to be introduced with the new vector formulation. We should check that gamma subtraction is working correctly for non-GP fits.

[bjfultn]

Can you resolve the conflicts

[sblunt]

Ofc. I'm still testing this branch so I'll resolve conflicts and let you know when it's ready to be merged.