Add a check for whether the spectrum is continuum subtracted for centroid

[eteq]

#270 (not merged at the time of this message, but probably will be soon) adds a centroid calculation. The trouble is that centroid algorithms can go wrong if there's a substantial un-subtracted continuum. @brechmos-stsci rightly points out in the docstring that this is essentially the user's responsibility... but we know how users can be about reading docs 😉.

So I think once #210 is in we should consider putting in a check at the top of the centroid function that tries to infer if the continuum has been subtracted. This could do something like:

1. run the continuum estimation function
2. ask if the resulting continuum is "close" (to be defined...) to zero.
3. If no, trigger a warning "Hey user, maybe you forgot to subtract the continuum? If so your centroiding is probably going to fail...", if yes, proceed with centroiding

Definitely will want an argument to turn this off though, since a user who explicitly says "no, trust me, it is subtracted" won't want to have to have the compute expense or potential annoyance of this warning.

[hcferguson]

Centroiding tends to be pretty robust to having continuum. Determining line widths is not. I'm on the fence about annoying users with a warning for centroiding. I do think it is worth it for widths, so maybe for consistency it's fine for centroiding. But I wouldn't want to add a big performance penalty in either case.

[nmearl]

This feature has been implemented in the function decorator that warns the user about potential non-reduced data being passed to operations. See #538.