[Checkpointing] Persist DataLoaderShard epoch counter across save/load_state

[s-zx]

What does this PR do?

Fixes #3996.

accelerator.save_state / accelerator.load_state previously round-tripped
samplers and (when use_stateful_dataloader=True) the dataloader state_dict,
but did not persist the per-epoch iteration counter on
DataLoaderShard / DataLoaderDispatcher.

That counter is what seeds SeedableRandomSampler on the next epoch (via
set_epoch() → sampler.set_epoch(epoch) → deterministic shuffle seed), so a
resumed run replayed the epoch-0 shuffle order instead of the correct epoch's
permutation. This reproduces exactly the "shuffle sequence replays from
iteration 0" behavior reported in the issue with use_seedable_sampler=True.

Fix

Extend save_accelerator_state / load_accelerator_state to write a small
dl_iteration.bin sibling file per dataloader when the dataloader exposes an
iteration attribute, and restore it on load via set_epoch so samplers and
IterableDataset.set_epoch hooks stay in sync.

• Save path is unconditional on hasattr(dataloader, "iteration") — so it also
  covers the non-stateful path (use_stateful_dataloader=False).
• Load path checks input_iteration_file.exists(), keeping backward
  compatibility with checkpoints produced by older accelerate versions.
• Restoration uses set_epoch(iteration) rather than direct assignment so
  downstream samplers / datasets receive the epoch change.

Tests

Added unit tests to tests/test_state_checkpointing.py:

• test_dataloader_iteration_counter_is_persisted — round-trips iteration through save/load.
• test_load_state_tolerates_missing_iteration_file — backward compatibility.

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

@muellerzr @BenjaminBossan