fix flaky test and add random seed to have deterministic tests

[patricefortin]

In some cases (non deterministic), test_filt was failing due to the strategy on PSD comparison.

The change here fix the test, and also add a random seed for which the test was failing.

[patricefortin]

This image shows the values of the sum of psd for the first frequencies, when the test is failing

plt.plot(np.arange(raw_psd.shape[1])[:10], np.sum(raw_psd, axis=0)[:10], label="raw")
plt.plot(np.arange(raw_filt_default_psd.shape[1])[:10], np.sum(raw_filt_default_psd, axis=0)[:10], label="filtered")
plt.legend()
plt.show()