There was a mismatch in the templates used internally in
bayesCBC and the one used in BayesWave to compute the residuals
(and likelihood). This commit updates the way the projection
is calculated (and this fixes the previous issue with parallel
tempering because now the likelihood is computed correctly..)