Optimize vector-number math by avoiding RealNumber_Check

[Starbuck5]

Take the code pygame.Vector2((10,20)) * 30. This operation is handled in the function vector_generic_math.

Before this PR, the code checks if PyFloat_AsDouble will work using the RealNumber_Check function, then it calls PyFloat_AsDouble. After this PR, it checks if PyFloat_AsDouble will work by calling PyFloat_AsDouble and handling the error if it fails. This avoids a call by us into RealNumber_Check (PyNumber_Check and !PyComplex_Check). I checked the Python source code, the code in PyNumber_Check is checking the same things PyFloat_AsDouble does, and PyComplex doesn't provide the right internal functions to be converted to a float/double.

Test results:

Code	Speedup
Vector2 * float	5.7%
Vector2.elementwise + float	5.8%

Results are quite noisy, especially since this is a relatively small optimization. I still it's worth checking in though!

Basic testing script

import pygame
import time
import random

l = [pygame.Vector2(random.random() * 100, random.random() * 100) for _ in range(200)]
le = [v.elementwise() for v in l]

start = time.time()
for _ in range(10000):
    for v in le:
        v + 3.956
print(time.time()-start)

start = time.time()
for _ in range(10000):
    for v in l:
        v * 3.956
print(time.time()-start)

[ankith26]

LGTM thanks! 🎉

[damusss]

Thank you for the optimization! 🎉 ❤️