Relational binops on unit type are nonsensical

[brson]

The following is currently what holds for relational binops on ()

  assert (() < ());
  assert (() <= ());
  assert (!(() > ()));
  assert (!(() >= ()));

Now, because we have relational binops over structural types, we have to either have relational binops over all types, or make structural comparisons much more complicated. But the implementation for unit is arbitrary. Suggest the following

  assert (!(() < ()));
  assert (() <= ());
  assert (!(() > ()));
  assert (() >= ());

[msullivan]

It looks like it just considers every relational operation over unit to be true.

[brson]

() > () and () >= () are false

[msullivan]

Oh. Oops. The cmp glue considers everything to be true, and there is cmp glue for ==, <, <=.

[brson]

So maybe the compare glue should consider < to be false.

[msullivan]

Closed by #663.