More to come. Namely, error context which changes throughout the program
(ie output to either stderr or to the command/status window) and
consolidating the debug code, as well.

Specific "modules" should include what they really need and not rely on
some other header including files.

Obviously, now, headers need to be explicitly included.

Might be better, in the future, to append -Wformat=2 to CFLAGS and also
use the printf format attribute.

Which, in my opinion, is a good thing because, as a consequence, there's
less pre-processor goo in the tree.

From now on, there's no need to specify the relative path of local
header files.

It might generate a SIGSEGV exception if it's accessed.
Caught by the new wrapper function.

Now, with error contexts, the execute function can, also, be called when
parsing the configuration file, without having to worry about where to
output its messages.

That is, when parsing the configuration file, the messages are printed to
standard error and, afterwards, they get painted to the command/status
window.

The fparseln(3) function already returns NULL on EOF. Check it and move
the ferror(3) call outside the loop.

It already gets printed by the error functions.

This uses the red-black tree implementation from OpenBSD. Other data
structures should probably be converted in the future, accordingly,
instead of rolling our own and sticking only to arrays.

Not working for other operating systems, yet.

From now on, when implementing a new e-command, there's no need to
duplicate code (ie call getopt, set flags and values). It's just a
matter of calling the arguments functions.

On other operating systems, the local tree.h file is included. Not
tested, though. The compat code probably needs some rework.