Rounding for symmetric

[dmcclean]

-- | Construct a symmetric interval.
--
-- >>> symmetric 3
-- -3 ... 3
symmetric :: Rounded TowardInf p -> Interval p
symmetric b = coerce (negate' b) ... b

But in intervals we had another test case:

-- >>> symmetric (-2)
-- -2 ... 2

I think both test cases belong, because that is (almost) the behavior of the definition for the rounded intervals too. If the negative input test case doesn't belong, then probably a test case saying that negative input is an exception does.

But I think this means that we need symmetric :: Rounded AwayFromZero p -> Interval p, right?