apply_sharding() check does not care about sharding priorities

🐛 Describe the bug

The following, in my opinion valid, snippet fails with

import torchdata.datapipes as dp
from torch.utils.data.datapipes.iter.sharding import SHARDING_PRIORITIES
from torchdata.dataloader2 import MultiProcessingReadingService, DataLoader2

pipe = dp.iter.IterableWrapper(range(10))
pipe = pipe.sharding_filter(SHARDING_PRIORITIES.DISTRIBUTED)
pipe = pipe.sharding_filter(SHARDING_PRIORITIES.MULTIPROCESSING)
    
rs = MultiProcessingReadingService(num_workers=1)
dl = DataLoader2(pipe, reading_service=rs)

for x in dl:
    print(x)

RuntimeError: Sharding twice on a single pipeline is likely unintended and will cause data loss. Sharding already applied to ShardingFilterIterDataPipe while trying to apply to ShardingFilterIterDataPipe

I.e. its currently not possible to first shard based on MPI rank, and then further shard based on (per-process) io worker rank, despite there being mechanisms built into sharding_filter for that purpose.

This check is overly restrictive in my opinion.

Versions

main branch & torch nightly.

Start by locating apply_sharding() and the sharding_filter implementation, then reproduce the provided DataLoader2 snippet with DISTRIBUTED followed by MULTIPROCESSING priorities. Trace how sharding priorities are checked; done means valid sequential priority-based sharding no longer raises the duplicate-sharding error and the relevant behavior is covered by tests.