FIX: Remove 0.0 from candidates in solve_discrete_riccati

[oyamad]

Fix #356

[oyamad]

test_robust_rule_vs_simple fails with a very small difference:

======================================================================
FAIL: test_robust_rule_vs_simple (quantecon.tests.test_robustlq.TestRBLQControl)
----------------------------------------------------------------------
Traceback (most recent call last):
 File "/home/travis/build/QuantEcon/QuantEcon.py/quantecon/tests/test_robustlq.py", line 82, in test_robust_rule_vs_simple
 assert_allclose(Kr_pf, Ks_pf, rtol=1e-4)
 File "/home/travis/miniconda/lib/python3.6/site-packages/numpy/testing/utils.py", line 1395, in assert_allclose
 verbose=verbose, header=header, equal_nan=equal_nan)
 File "/home/travis/miniconda/lib/python3.6/site-packages/numpy/testing/utils.py", line 778, in assert_array_compare
 raise AssertionError(msg)
AssertionError:
Not equal to tolerance rtol=0.0001, atol=0
(mismatch 66.66666666666666%)
 x: array([[ -4.332446e-25, -8.620690e+01, -3.903128e-17]])
 y: array([[ -3.704241e-25, -8.620690e+01, -3.003457e-17]])

x = np.array([[ -4.332446e-25, -8.620690e+01, -3.903128e-17]])
y = np.array([[ -3.704241e-25, -8.620690e+01, -3.003457e-17]])
x - y

array([[ -6.28205000e-26,   0.00000000e+00,  -8.99671000e-18]])

I am going to set atol to a small positive value.

[coveralls]

Coverage increased (+0.02%) to 94.79% when pulling 6aea839 on fix-riccati into ea900c4 on master.