slight modification for HSIC speedup

[MarkDana]

Just some speedup on matrix multiplication (e.g., multiplied by all ones matrix). 5-10 times faster (depends on sample size).

X = np.random.uniform(0, 1, (15000,))
Y = X ** 2 + np.random.uniform(0, 1, (15000,))
tic = time.time()
test_stat, p = hsic_test_gamma(X, Y)
print(f'now used: {time.time() - tic: .5f}s')

from causallearn.search.FCMBased.lingam.hsic import hsic_test_gamma as hsic_test_gamma_old
tic = time.time()
test_stat_old, p_old = hsic_test_gamma_old(X, Y)
print(f'originally used: {time.time() - tic: .5f}s')

assert np.isclose(test_stat, test_stat_old)
assert np.isclose(p, p_old)
print('equivalent test passed.')

'''
now used:  6.78904s
originally used:  65.28648s
equivalent test passed.
'''