Best sample selection uses different confidence metrics in writer vs. analyze #127

In src.boltzgen.task.predict.writer:write_on_batch_end, the “best” sample in the batch is determined according to

confidence = 0.8 * pred_dict["iptm"] + 0.2 * pred_dict["ptm"]

However, in src.boltzgen.task.analyze.analyze_utils.get_best_folding_sample(), the “best” sample in the batch is determined according to

confidence = 0.8 * folded["design_to_target_iptm"] + 0.2 * folded["design_ptm"]

This often results in a different refolded sample being written to the output, as compared to the sample that is used for the analysis/filtering step.