Check 'thread' on Windows

[jgillis]

Discussed in #3142

[jgillis]

Test script:

from casadi import *
import os


n_threads = os.cpu_count()//2

N = 1000

x = MX.sym("x",N)

y = sin(x)

for i in range(10000):
    y = sin(y)

f = Function("f",[x],[y])

import time


results = {}
for type in ["serial","thread"]:

    F = f.map(n_threads,type)

    X_ = np.random.random((N,n_threads*N))
    Y_ = np.zeros((N,n_threads*N))
   
    # Use buffered evaluation to avoid overhead

    [buf,F_eval] = F.buffer()
    buf.set_arg(0, memoryview(X_))
    buf.set_res(0, memoryview(Y_))

    t0 = time.time()
    F_eval()
    dt = time.time()-t0
    print("Time ", type, ": " , dt)
    
    results[type] = dt

print("Speedup: ", results["serial"]/results["thread"], " - expected :", n_threads)