RGB PNG output looks strange

[m-mohr]

I'm trying to create a RGB PNG, which looks good as GTiif:

datacube2 = connection.load_collection(collection_id = "SENTINEL2_L2A_MOSAIC_120", bands = ["B04", "B03", "B02"], spatial_extent = {"east": 9.722042179836016, "north": 43.14486158000392, "south": 41.34576268423038, "west": 8.44096604833396}, temporal_extent = ["2020-04-01T00:00:00Z", "2020-04-30T00:00:00Z"])

def reducer1(data, context = None):
    data1 = process("mean", data = data)
    return data1

datacube4 = datacube2.reduce_dimension(dimension = "t", reducer = reducer1)
datacube5 = datacube4.save_result(format = "PNG")

The resulting image looks weird:

Some observations from this:

1. I suspect I still need to do a linear scaling to 0 to 255, but the metadata doesn't show the min/max values for the bands. I'd propose to add statistics to raster:bands in the stac metadata, otherwise it's hard to know how to apply the scaling.
2. If you try to write a PNG and the values do exceed the allowed values of a PNG (e.g. < 0 or > 255), I would hope to find a warning in the job logs.