Finalize the pygmt.which wrapper

[seisman]

Upstream documentation: http://docs.generic-mapping-tools.org/dev/gmtwhich.html
Upstream long option: https://github.com/GenericMappingTools/gmt/blob/master/src/longopt/gmtwhich_inc.h

The gmtwhich module is one of the simplest modules and has very few options:

-A: Only consider files that the user has permission to read [Default consider all files found].

I think this option is useful and should be implemented. GMT's long option name readable sound good.

-C: Instead of reporting the paths, print the confirmation Y if the file is found and N if it is not.

Currently, we raise FileNotFoundError exception if a file is not found. Sometimes, we may want to take specific actions if a file is not found, so this option is still useful. GMT's long option name is confirm and return Y/N. I think it makes little sense in Python. Instead, we can return the full path if found and None if not found? Or True/False?

-D: Instead of reporting the paths, print the directories that contains the files.

Useless. In Python, we can always use os.path.dirname or Path().parent instead.

-G[a|c|l|u]: download the file if it's a downloadable file

Already implemented but is not Pythonic. The single-letter argument (e.g., download="a") is not Pythonic. More Pythonic arguments are True, auto, cache, local, user. GMT uses user, cache, current, data instead. I feel my suggestions make more sense and we can ask for upstrem changes.

-V: verbose level.

Common option. Already implemented.

We also need to decide:

• if we want to use virtualfiles for output [xref Get rid of temporary files from pygmt functions and plotting methods #2730]
• if output type file/numpy/pandas should be supported [xref Consistent table-like output for PyGMT functions/methods #1318]