Indexing tree should create new tree

[TomNicholas]

Inspired by this example in the stackstac documentation

lowcloud = stack[stack["eo:cloud_cover"] < 20]

we should ensure that you can index a datatree with another (isomorphic) datatree, so that the above operation would work even if stack is a DataTree instance.

This is another map_over_subtree-type operation, but it needs careful testing because the __getitem__ function in xarray objects already does so many different things. This won't work with the code as-is because at the moment the DataTree naively dispatches the __getitem__ call down to the wrapped dataset.

datatree/datatree/datatree.py

Line 238 in cd06951

return self.ds[key]

[TomNicholas]

To clarify, in order for this to work several things need to happen:

1. stack["eo:cloud_cover"] needs to realise that "eo:cloud_cover" is not a tree, not a group in the tree, but a variable name. Then it needs to select the "eo:cloud_cover" variable from all nodes in the subtree, and return a tree containing only those variables. That in itself requires something like Ignore missing dims when mapping over tree #67 but ignoring nodes for which that variable is not present, at least for deeply-nested trees...
2. stack["eo:cloud_cover"] < 20 needs to perform this comparison node-wise, returning a tree of results (hopefully this should already work...
3. stack[stack["eo:cloud_cover"] < 20] needs to use the tree passed to perform a node-wise indexing operation, returning a new tree. (Or we could just .where)

Basically this is a really complicated usage example because it uses multiple different code-paths within __getitem__ sequentially within one line of user code 😅

[TomNicholas]

Closing in favour of pydata/xarray#9341