Implement support for Thompson sampling

[kiudee]

The acquisition functions are currently called using mu and std already evaluated for a set of points:

bayes-skopt/bask/acquisition.py

Lines 24 to 28 in 071d504

	mu, std = gpr.predict(X, return_std=True)
	for j, acq in enumerate(acquisition_functions):
	tmp_out = acq(mu, std, **kwargs)
	if np.all(np.isfinite(tmp_out)):
	acq_output[j] += tmp_out

For Thompson sampling we instead want to sample a GP instead of the mean process.