Simplify refinements whenever possible

[sgpthomas]

If you have 0 <= V and V <= 0 in your set of qualifiers, then you can infer refinements of the shape 0 <=V and V <= 0. We should use an SMT solver to simplify this into V == 0 even if V == 0 isn't in the set of qualifiers.

This will just make the inferred refinement types easier to read.

[dijkstracula]

Weirdly, z3's simplify nor some straightforward tactics don't seem to do the needful here: I would have thought it'd be smart enough to figure it out, but I guess I'm misunderstanding its limits or just driving it wrong:

>>> V = Int("V")
>>> simplify(And(0 <= V, V <= 0))
And(V >= 0, V <= 0)
>>>
>>> t = Tactic('solve-eqs')
>>> t(And(0 <= V, V <= 0))
[[V >= 0, V <= 0]]
>>>

Rubbing everyone's favourite CS 386L tactic on the result yields a funny result, too...

>>> t = Tactic('lia')
>>> t(And(0 <= V, V <= 0))
[[]]
>>>

Broadly, though, it's interesting to think about what sequence of tactics might be useful for our domain here. (word to the wise: z3.describe_tactics() will print out all the available ones in the REPL.)