Correct nominal response for unaccounted residual delays

This MR changes the nominal response in uncertainty.py to include the MAP value from the MCMC when sampling from the MCMC for residual time delays. The reason to do this is because the normalization of GPR measurements do not use the residual time delay MAP values (assume residual delay is zero) and FIR filter generation also (I think; to be confirmed) does not use any residual delay when creating the FIR filters. Thus, the GPR should cover the incorrect value and we don't need to double-count it from the MCMC, so we should remove it.

Test before this MR with only the sensing MCMC chain but no other MCMC or GPR:

Test using this MR with only the sensing MCMC chain but no other MCMC or GPR:

This verifies that the MCMC residual delay should not be sampled with its offset, as shown by the second plot where the phase drop at higher frequencies is now gone.