ignore index included in foreground when calculating dataset fingerprint

I noticed odd foreground_intensities for my dataset.

I have this dataset.json

{
  "channel_names": {
    "0": "CT"
  },
  "labels": {
    "background": 0,
    "sm": 1,
    "sat": 2,
    "vat": 3,
    "other_tissue": 4,
    "ignore": 5
  },
  "numTraining": 3056,
  "file_ending": ".nii.gz"
}

Preprocessing gives the following stats:

foreground_intensity_properties_per_channel": {
        "0": {
            "max": 10623.0,
            "mean": -321.3395918581319,
            "median": -66.0,
            "min": -3040.0,
            "percentile_00_5": -3024.0,
            "percentile_99_5": 701.0,
            "std": 790.7851868216073
        }
    },

When calculating manually I got much more reasonable values (99% -160HU to 100HU, mean ~-40HU). I figured it might be that nnunet was including the ignore class when calculating the statistics. Looking at DatasetFingerprintExtractor.collect_foreground_intensities I see foreground_mask = segmentation[0] > 0, which I think confirms my suspicion.

Fix: update foreground mask to properly exclude the ignore class (if present)