Refactor FITS table docs #3782

I receive a whole category of FAQs, similar to spacetelescope/PyFITS#101 (no reason I'm picking on it specifically other than that it's the most recent example), that go something like: "I tried to create a FITS table by creating a bunch of Column objects, and something or other didn't work."

The problem is that the current docs for creating FITS tables, much of which is inherited from earlier versions of PyFITS that were less flexible, emphasize a workflow that requires a lot of direct fussing with the FITS format: One has to specify column formats using the old FORTRAN-like format codes, and often involving other quirks like getting setting the correct TDIMn keywords for multidimensional columns, and just generally more fussing than is usually really necessary.

These interfaces should still be documented, as they are useful for cases where the user needs very fine control over how the FITS format is written. But in most cases this is not necessary, and what PyFITS does automatically is completely sensible. So the docs should first emphasize how tables can be created from Numpy structured arrays, or by using the Astropy Table interface (this is already brought up in #3552, but we need to go even further than that).

To go even further, there might be some room for improving the the API on this front. For example, rather than manually creating Column objects it should be good enough, in most cases, to pass in a list of arrays, along with maybe a list of column names (similar to the existing interface for creating Table objects).