Decide on correct meta copying behaviour in rebin

[Cadair]

Describe the bug

Pre #455 we didn't copy the meta attribute at all if it wasn't rebin aware:

ndcube/ndcube/ndcube.py

Line 1220 in fb788f7

new_meta = self.meta

post #455 we deepcopy it:

ndcube/ndcube/ndcube.py

Line 1280 in badee33

new_meta = deepcopy(self.meta)

We had a long discussion about this in the matrix channel, but to summarise:

• Our historical policy was that we made no assumptions about .meta so it could be any object of any arbitrary complexity. I feel like defaulting to deepcopy is not in the spirit of this policy.
• It's safer to copy than not deepcopy because a deepcopy could use a lot of memory if a large array is in the .meta, and it's safer and easier for someone to explicitly deepcopy afterwards than it is to undo a deepcopy.
• Users may well be surprised by the fact that the whole .meta is the same or the values of a dict as .meta are the same if we don't copy (old behaviour) or only copy.

[samaloney]

So if the meta is axis-aware or not I will get different behaviour? If it is not axis-aware meta will be a reference to the original cubes meta but if it is axis aware it will be a new rebinded meta? My initial thought is a rebinded cube is a completely new cube and that it would not share anything with the original cube and if I changed the meta on the rebinded cube I would not expect it to alter the meta on the orig cube 🤯 I can see times where sharing it would be beneficial or even necessary.

I think there are cases where either is correct so maybe add an kwarg to control and default to old behaviour. In any case I think it needs to be called out in the documentation as a potential footgun.

[Cadair]

If it is not axis-aware meta will be a reference to the original cubes meta but if it is axis aware it will be a new rebinded meta?

I am proposing to make it copy not deepcopy for what it's worth.