Refactor async datatree loading to reuse open_groups_as_dict_async

[aladinor]

Summary

Follow-up from PR #10742 (async DataTree loading optimization). As suggested by @keewis in the PR review:

logically I think xr.open_datatree(...) == xr.DataTree.from_dict(xr.open_groups(...)), so it might make sense to have _open_datatree_async call _open_groups_as_dict_async

Current State

After #10742, the zarr backend has two similar async implementations:

• _open_datatree_from_stores_async() - used by open_datatree()
• open_groups_as_dict_async() - used by open_groups()

Both do similar work but with some differences:

• Semaphore usage for concurrency control
• Index creation handling

Proposed Change

Refactor _open_datatree_from_stores_async to internally call open_groups_as_dict_async, reducing code duplication and ensuring consistent behavior between the two APIs.

[AAlexxis222]

Hi! I would like to tackle this refactor.

I recently finished a widespread refactor in arviz-plots (standardizing signatures across 34 modules) and I am looking to contribute to Xarray as part of my GSoC preparation.

I will start by analyzing the implementation differences between _open_datatree_from_stores_async and open_groups_as_dict_async to plan how to reuse the latter without breaking the existing logic.