Get rid of make conj

[hferee]

Try to replace make_conj with And (and similarly for disjunction and implication) to see if it makes the benchmarks worse

[hferee]

Reporting progress: while investigating #33, I changed the implementation of choose_conj with its naive implementation ; thus impacting make_conj and conjunction.

This makes for identical results (and better speed) in all benchmark cases, except for (x <-> (p ∨ ¬p)) ∧ (y <-> (q ∨ ¬q)) ∧ ¬(p ∧ q) which times out.
The issue is in conjunction ; seemingly highly inefficient in that case.
Counterintuitively, it seems to make more calls to Provable_dec.

We'll see if the solution above fixes this issue too.