general polynomial detrending

[petigura]

@awhoward, @bjfultn, @sblunt,

Based on @awhoward's question about the decorrelation fitting, I think we should start a discussion about implementing a likelihood class that can incorporate polynomial decorrelation an arbitrary string of independent variables e.g. svals, halpha, etc. This is low priority, and I don't think we should stress about getting this in v1.0.X

# define params as before, but with
params['c0'].value = 0
params['c1'].value = 1
like = PolyDecorLikelihood(t, y, yerr, x, xerr, prefix='c', deg=1)

Then instead of the standard likelihood calc

nresid = (y - model)/yerr
_logprob = 0.5 * np.sum((nresid)**2)

we would use:

decor = [params['c0'],params['c1']]
nresid = (y - model - polyval(decor,x))/yerr
_logprob = 0.5 * np.sum((nresid)**2)

[petigura]

I can take the lead on implementing this since I need to do this for my K2-24 project.

[bjfultn]

When you get to this, lets repurpose the radvel.plotting.correlation_plot function to plot the correlations.