Should the io machinery use astropy's "config-finding" machinery?

The current machinery for finding custom I/O plugins involves finding a ".specutils" directory in the users home directory. This works for now, but has some problems: it's unnatural to some in windows, it's not always portable enough for some more obscure environments, it doesn't let the user control where specutils puts things, etc.

Astropy has gotten around this through a sophisticated mechanism of letting environment variables find the correct location. One could argue it's *too* sophisticated of course, but at least sharing that with astropy would make it only one learning curve.

Alternatively, we could re-think this entirely and require use of entry_points or similar (which has its own trade-offs).

Start by reviewing the current custom I/O plugin discovery that searches for a .specutils directory in the user's home directory, then compare it with Astropy's config-finding machinery and the entry_points alternative. Done should mean choosing and implementing a supported discovery approach that addresses Windows, portability, and user-controlled locations.

Written by the indexing model from the issue text.