BUG: Fix bug for input with int elements

[oyamad]

I found the same bug as in the Julia version. For an input matrix with integer elements, gth_solve returns a wrong answer:

>>> P = [[-3, 3], [4, -4]]  # Transition rate matrix
>>> gth_solve(P)
array([ 0.5,  0.5])

The correct answer is [4/7, 3/7].
(My choice of the instance was not clever so that the test failed to detect this bug...)

The current code now converts the input to float.

[coveralls]

Coverage remained the same when pulling 493ab6e on oyamad:gth_solve into 42b4344 on QuantEcon:master.

[jstac]

@oyamad Well spotted, many thanks.