Create a full production ready multiscales pipeline

As part of their blog post Earthmover put together a nice reference implementation of multiscale generation https://github.com/earth-mover/icechunk-multiscales-demo. It is great as a reference but has a few limitations that don't make it suitable for usage with operational NASA datasets.

• Coiled. Though we can use Coiled for quick cluster scaling experimentation it is unlikely to be adopted as a long term cluster solution for NASA sow we'll need to adapt this to use our own managed Batch clusters or potentially a Lambda approach depending on compute and memory requirements.
• The demo approach is static and operational datasets will require hooks to a new chunk notification system that broadcasts notifications of chunk additions or updates to the native resolution arrays (or more likely virtual arrays) so that the multiscales pipeline can update regions accordingly.
• We should include the ability to incrementally populate multiscale arrays on-demand based on user requests for less used regions / variables. See https://github.com/NASA-IMPACT/veda-odd/issues/427 for more details on this approach.

This is critical for visualization performance and wider adoption of virtualization as a basis for a more modern data system.

Start by reviewing the reference implementation in https://github.com/earth-mover/icechunk-multiscales-demo and the on-demand approach described in issue #427. Define the operational architecture for managed Batch clusters or Lambda, chunk-addition and update notifications, and incremental population of requested regions and variables; done means these requirements are supported for NASA datasets.