Bug in integrality in IC non-linears

Hi,

The semantics of the `#=', etc. variants of the IC constraints is that the
variables appearing in such constraints, except where it is inappropriate,
should be constrained to be integral.  This is not happening.

E.g.

[eclipse 7]: [X, Y] :: 1..10, X #= K * Y.

X = X{1 .. 10}
K = K{0.099999999999999978 .. 10.000000000000004}
Y = Y{1 .. 10}


Delayed goals:
	ic : (-(_492{1 .. 10}) + X{1 .. 10} =:= 0)
	ic : (_492{1 .. 10} =:= K{0.099999999999999978 .. 10.000000000000004} * Y{1 .. 10})
	ic : (K{0.099999999999999978 .. 10.000000000000004} =:= _492{1 .. 10} / Y{1 .. 10})
	ic : (Y{1 .. 10} =:= _492{1 .. 10} / K{0.099999999999999978 .. 10.000000000000004})

K should be integral (particularly since if Y is ground at runtime, K is
constrained to be integral).

Cheers,
Warwick