Skip chunk coordinate enumeration in resize when array is only growing (#3650)

[jakenotjay]

Motivation
Array.resize() enumerates all chunk coordinates if delete_outside_chunks is True (its default value), these old and new shapes are added into Python set objects to compute which chunks to delete. #3650 notes that this is totally unbounded in memory - even if the array is only growing and no chunks need deletion.

In our case our array has approximately ~220 million chunks, and we want to add a new time step. Building these two sets of tuples explodes memory to >20GB and does not complete (at least deployed in a no-swap environment, I never left my laptop running long enough). Furthermore, the set difference will always be empty because growing a dimension can't produce chunks outside the new shape.

Fix
Rather than always checking which is O(total_chunks), first check if the new_shape is >= the dimensions of the old shape. If array is only growing, skip enumeration.

This is a targeted mitigation, rather than a complete solution to the problem described by 3650, a more complete solution would construct outside chunk coords from the shape diff, rather than naively enumerating all coords.

TODO:

• Add unit tests and/or doctests in docstrings
• N/A Add docstrings and API docs for any new/modified user-facing classes and functions
• N/A New/modified features documented in docs/user-guide/*.md
• Changes documented as a new file in changes/
• GitHub Actions have all passed
• Test coverage is 100% (Codecov passes)

[rabernat]

Great fix, but it needs a test, no?