TrueContinuum var_lss is computed in a grid that is too fine

[iprafols]

Following the changes made in PR #994 I observed that the values of sigma_lss for the TrueContinuum expected flux are not stable (see red vs blue line in the plot). I investigated the origin of this change and I found out that in the TrueContinuum class, sigma_lss is computed in the Forest.log_lambda_grid grid and then interpolated to the self.log_lambda_var_func_grid grid before saving. This "native" grid is shown as the orange line in the plot. The change observed in the VAR_STATS HDU is associated with the noisy nature of this function. This might be an issue for the runs using TrueContinuum expected flux but is outside the scope of this PR (I will now file this as an issue), but I don't know how important it will be.

To solve this issue, I suggest smoothing the "native" grid right after it is computed to reduce the impact of noise