cmdstan: `generate_quantities` handles `save_warmup=True` "wrong".

Summary:

If save_warmup=True, generate_quantities takes the first num_samples warmup and samples to generate the gqs.

https://github.com/stan-dev/cmdstan/blob/develop/src/cmdstan/command.hpp

    size_t num_cols = param_names.size();
    size_t num_rows = fitted_params.metadata.num_samples;
    // check that all parameter names are in sample, in order
    if (num_cols + hmc_fixed_cols > fitted_params.header.size()) {
      msg << "Mismatch between model and fitted_parameters csv file \"" << fname
          << "\"" << std::endl;
      throw std::invalid_argument(msg.str());
    }
    for (size_t i = 0; i < num_cols; ++i) {
      if (param_names[i].compare(fitted_params.header[i + hmc_fixed_cols])
          != 0) {
        msg << "Mismatch between model and fitted_parameters csv file \""
            << fname << "\"" << std::endl;
        throw std::invalid_argument(msg.str());
      }
    }
    return_code = stan::services::standalone_generate(
        model,
        fitted_params.samples.block(0, hmc_fixed_cols, num_rows, num_cols),
        random_seed, interrupt, logger, sample_writer);

Has to be changed into something that handles this differently, also incorporating the thinning parameter.

Current Version:

v2.26.1