Also do some minor clean up.

This reverts commit 24fa20e5f8a2042caa19f1f6505102c5434cce54.

Although these symmetric variants sometimes look "better", they
are really annoying and should IMHO never be used:

1.) For lemmas there is now a choice between >= and <=. Since there is
no longer a canonical choice, it is very easy to introduce a lot of
inconsistencies in statements of lemmas.

2.) For automation the situation becomes annoying, you have to built in
stuff for both >= and <=. That is very error-prone.

3.) For N and Z the notions x <= y and y >= x are not even convertible!
That means that done/by does not solve x <= y if you have y >= x and if
avoids you applying certain lemmas.

And use more uniform variable names.

This one (previously solve_elem_of) was hardly used. The tactic that uses
naive_solver (previously esolve_elem_of, now solve_elem_of) has been
extended with flags to say which hypotheses should be cleared/kept.

These are hardly used, and confusing since we have so many operations of
different arities that distribute.

These are unused and not very useful anymore now that we have gmap.

Also, use a different encoding of lists.