Description:
------------
The segfaults are caused by precision loss of large longs being converted to doubles when the `step` parameter is a double. The for loops continue infinitely since the `step` being added/subtracted upon each iteration is too small to be represented accurately as a double.

The attached patch adds another condition to each of the for loops by ensuring that the number of iterations in the loop are less than the size of the range (`__calc_size`). This prevents both loops from continuing until segfaulting, however doesn't give an accurate result (due to `__calc_size` being incorrect because of aforementioned precision loss).

Test script:
---------------
var_dump(count(range(PHP_INT_MIN, PHP_INT_MIN + 513, .01)));
var_dump(count(range(PHP_INT_MIN + 513, PHP_INT_MIN, .01)));

Expected result:
----------------
// A completely correct output would be:
int(51400)
int(51400)

Actual result:
--------------
Segmentation fault: 11
Segmentation fault: 11