Improve WCS handling when slicing/extracting regions

[rosteen]

As of #918, slicing the spectral axis of a Spectrum1D will drop the existing WCS and recreate a fresh lookup table GWCS from the new spectral axis. It seems like it would be better to propagate any spatial components of the existing WCS for a multi-dimensional Spectrum1D. Additionally, for spatial slicing it would be nice to properly handle any spatial components of the WCS so that the resulting Spectrum1D has a WCS with as much information as the original. I'm not sure, but NDCube may handle this better than we currently do here for at least spatial dimensions - I'll have to check. It may be that we're overriding something useful.

[eteq]

As referenced in #918 (comment), we may want to tack onto a PR solving this something along the lines of "clearly document the WCS behavior when slicing/extracting". E.g., find some place in the narrative docs to explain the current behavior (and the fact that the "old" wcs is preserved in meta a la #918)