[FEATURE] Aggregator should write per-instance results

[Hendrik-code]

Is your feature request related to a problem? Please describe. The aggregator writes one evaluation line for each sample, not instance in that sample. This means that afterwards, one cannot know the distribution of those instance-wise evaluations within one sample

Describe the solution you'd like The aggregator should write one line in the tsv for each instance in a sample. Then we can create functions in the panoptica_statistics class that make a post-evaluation analysis (for example, does one instance dominate the global_bin_dice score, ...)