Wrong start_time, end_time attributes after MultiScene.blend(blend_function=timeseries)

[gerritholl]

Describe the bug

After producing a scene with MultiScene.blend(blend_function=timeseries), the attributes start_time and/or end_time on the resulting Scene are wrong.

To Reproduce

import hdf5plugin
from glob import glob
from satpy.multiscene import MultiScene, timeseries
from satpy.utils import debug_on; debug_on()

ms = MultiScene.from_files(
        glob("/media/nas/x21308/MTG_test_data/2022_05_MTG_Testdata/RC010[012]/*BODY*.nc"),
        reader="fci_l1c_nc",
        group_keys=["repeat_cycle_in_day"])
channels = ["ir_105", "ir_133", "wv_63", "wv_73"]
ms.load(channels)
sc = ms.blend(blend_function=timeseries)
print(sc.start_time, sc.end_time)
print(ms.scenes[0].start_time, ms.scenes[-1].end_time)

Expected behavior

I expect as the final output from the print functions:

2017-09-20 16:30:08 2017-09-20 16:59:22
2017-09-20 16:30:08 2017-09-20 16:59:22

Actual results

The result of the print functions is:

2017-09-20 16:40:08 2017-09-20 16:49:22
2017-09-20 16:30:08 2017-09-20 16:59:22

Environment Info:

• OS: openSUSE 15.3
• Satpy Version: v0.41.1-18-gb92fe9ebb (current GitHub main)

Additional context

The attributes are also wrong on the individual datasets that have been blended.

[djhoese]

@adybbroe and I had a conversation similar to this when he added his recent weighted blending. Right now the timeseries functionality uses combine_metadata which for some reason that I don't recall does not handle start/end time in any special way and only averages them.

Usage:

satpy/satpy/multiscene.py

Line 124 in 9c960e5

res.attrs = combine_metadata(*[x.attrs for x in expanded_ds])

Definition docstring:

satpy/satpy/dataset/metadata.py

Lines 35 to 38 in 9c960e5

	the returned dictionary. By default any keys with the word 'time'
	in them and consisting of datetime objects will be averaged. This
	is to handle cases where data were observed at almost the same time
	but not exactly. In the interest of time, lazy arrays are compared by

I think we need to update combine_metadata to min/max start/end times. @mraspaud @pnuu?