PEtab pipeline: Only compute sensitivities with respect to optimized parameters

[dweindl]

... currently this is done w.r.t. all parameters.

While less performance critical with ASA, it might still result in unnecessary simulation failures due to issues during gradient computation.

Related: #190, #343

[dweindl]

After #388, this will be less of an issue, since the more recent amici-petab-import will already treat petab-estimate=0 parameters as fixed parameters. What remains, is the handling of parameters w.r.t. which we only require sensitivity for certain simulation conditions. This then condition-specific amici-plist 1) needs to be written to the input files and 2) must be accounted for when aggregating the gradients from individual model simulations to the objective gradient.