Only merge in open_run(..., data='all') if raw is not a subset of proc

[philsmt]

open_run(..., data='all') throws an exception since #208 if raw is completely a subset of proc, because it will try to access the metadata of raw \ proc while merging this into proc. If this set is empty however, the call to DataCollection.run_metadata() fails due to access to files[0].

While this begs a larger question of whether such DataCollection objects such exist in the first place, @takluyver and me agreed to discuss this separately in #235. It is clearly not necessary and thus a bug in open_run do try this in the first place if the DataCollection is empty anyway.

[philsmt]

Now technically this only looks at sources, so if proc adds keys, but does not contain those in raw, it would still exclude the ones on raw and only show the added ones.

For AGIPD, data is actually replaced by calibration. Do we know this is the case for all detectors?

[takluyver]

I think the offline correction for all detectors uses the same source & key name as the raw data, so effectively replacing data.

EXtra-data currently assumes that each source exists in a single sequence of files, so we can't easily break it down to 'these keys from that file, those keys from another file'.

[philsmt]

Ah, this might've been the reason we sorted only by source in the first place. Thanks!

[takluyver]

LGTM