Move the`ground_truth_model` conditional out of `compute_poly_metric`

[jeipollack]

Description:

Currently, compute_poly_metric (and potentially other metric functions in metrics.py) contains conditional logic to handle the presence or absence of ground_truth_model. Even if ground_truth_model is "present" in the dataset, the ground truth model class is instantiated anyways, and repeatedly in each evaluate_<name_of_metric> call. This mixes configuration handling with metric computation, causes redundancy, and makes the code harder to maintain and test.

Proposed Options:

Move the conditional check out of compute_poly_metric (and other metrics) into one of the following:

1. MetricsParamsHandler.evaluate_metrics_polychromatic_lowres
2. A pre-validation step during MetricsParamsHandler.__init__, or a dedicated helper method like prepare_ground_truth_model

Preferred Approach:

Option 2 – handle the ground truth model setup in the initialization/pre-validation phase. This ensures that all metrics functions can safely assume a valid ground_truth_model and simplifies downstream logic.

Notes:

• This change would not alter functionality; it’s a refactor to separate configuration validation from metric computation.
• Existing tests (or new unit tests like test_missing_ground_truth_model_raises in Explore metrics_interface behavior under different evaluation flag configurations #186) can confirm that missing ground_truth_model raises an appropriate error.