`assert_float` is always passed when expected value and/or actual value are
infinity or NaN. This behavior seems unintentional.

Before this patch:

    assert_float(Float::INFINITY, 1.0)  #=> pass
    assert_float(-Float::INFINITY, 1)   #=> pass
    assert_float(1, 1/0)                #=> pass
    assert_float(1, -1/0)               #=> pass
    assert_float(1.0, Float::NAN)       #=> pass
    assert_float(Float::NAN, 1)         #=> pass

After this patch:

    assert_float(Float::INFINITY, 1.0)  #=> fail: Expected 1.0 to be Infinity.
    assert_float(-Float::INFINITY, 1)   #=> fail: Expected 1 to be -Infinity.
    assert_float(1, 1/0)                #=> fail: Expected Infinity to be 1.
    assert_float(1, -1/0)               #=> fail: Expected -Infinity to be 1.
    assert_float(1.0, Float::NAN)       #=> fail: Expected NaN to be 1.0.
    assert_float(Float::NAN, 1)         #=> fail: Expected 1 to be NaN.