Optimise math.ceil for known exact float

This matches a similar optimisation done for math.floor in
python#21072

Before:
```
λ ./python.exe -m timeit -r 11 -s 'from math import ceil' -s 'x=3.14' 'ceil(x)'
20000000 loops, best of 11: 13.3 nsec per loop
λ ./python.exe -m timeit -r 11 -s 'from math import ceil' -s 'x=0.0' 'ceil(x)'
20000000 loops, best of 11: 13.3 nsec per loop
λ ./python.exe -m timeit -r 11 -s 'from math import ceil' -s 'x=-3.14E32' 'ceil(x)'
10000000 loops, best of 11: 35.3 nsec per loop
λ ./python.exe -m timeit -r 11 -s 'from math import ceil' -s 'x=-323452345.14' 'ceil(x)'
10000000 loops, best of 11: 21.8 nsec per loop
```

After:
```
λ ./python.exe -m timeit -r 11 -s 'from math import ceil' -s 'x=3.14' 'ceil(x)'
20000000 loops, best of 11: 11.8 nsec per loop
λ ./python.exe -m timeit -r 11 -s 'from math import ceil' -s 'x=0.0' 'ceil(x)'
20000000 loops, best of 11: 11.7 nsec per loop
λ ./python.exe -m timeit -r 11 -s 'from math import ceil' -s 'x=-3.14E32' 'ceil(x)'
10000000 loops, best of 11: 32.7 nsec per loop
λ ./python.exe -m timeit -r 11 -s 'from math import ceil' -s 'x=-323452345.14' 'ceil(x)'
10000000 loops, best of 11: 20.1 nsec per loop
```

Modules/mathmodule.c

@@ -1125,8 +1125,13 @@ static PyObject *
 math_ceil(PyObject *module, PyObject *number)
 /*[clinic end generated code: output=6c3b8a78bc201c67 input=2725352806399cab]*/
 {
+ double x;
 
- if (!PyFloat_CheckExact(number)) {
+ if (PyFloat_CheckExact(number)) {
+ x = PyFloat_AS_DOUBLE(number);
+ }
+ else
+ {
  math_module_state *state = get_math_module_state(module);
  PyObject *method = _PyObject_LookupSpecial(number, state->str___ceil__);
  if (method != NULL) {
@@ -1136,11 +1141,10 @@ math_ceil(PyObject *module, PyObject *number)
  }
  if (PyErr_Occurred())
  return NULL;
+ x = PyFloat_AsDouble(number);
+ if (x == -1.0 && PyErr_Occurred())
+ return NULL;
  }
- double x = PyFloat_AsDouble(number);
- if (x == -1.0 && PyErr_Occurred())
- return NULL;
-
  return PyLong_FromDouble(ceil(x));
 }