Evaluator + `n_repeats > 1` + initial NaN in `tendency_of_total_water_path_due_to_advection` leads to NaN channel-mean RMSE

[jpdunc23]

When n_repeats > 1, the initial all-NaN timepoint of tendency_of_total_water_path_due_to_advection is appended to the end and results in NaN for the time-mean tendency_of_total_water_path_due_to_advection.

With n_repeats = 1, the following line avoids the initial NaN timepoint in tendency_of_total_water_path_due_to_advection: https://github.com/ai2cm/full-model/blob/2319dcee2e32213ffcbcfed7453a9b6dce14ac2f/fme/core/derived_variables.py#L87

[jpdunc23]

First noticed this issue in the following coupled run, but this should also happen in uncoupled atmosphere runs with n_repeats > 1.