Refactor `assign_files_to_rank` into a Scheduler Class

I’m thinking we should refactor [`assign_files_to_rank` ](https://github.com/Imageomics/hpc-inference/blob/f992676172728237d95b042d6ca27e698f5afa4d/src/hpc_inference/utils/distributed.py#L16) into a proper `Scheduler` class.

Right now, it's just a standalone function, but wrapping it in a class would give us better state management, easier testing, and make it way simpler to plug in new file distribution algorithms later.