Fail fast on empty action shuffle blocks

[fallintoplace]

Summary

• raise a clear error when the action shuffle dataset has no shuffle blocks
• add a regression test covering the empty-block case without pulling in the heavy dataset package imports

Root cause

ActionIterableShuffleDataset.__iter__() enters an infinite while True loop even when get_shuffle_blocks() returns an empty list. In that case torch.randperm(0) produces no indices, so the iterator yields nothing, increments epochs forever, and can leave workers hanging without a useful failure signal.

Validation

• uvx ruff check /Users/hoangvu/Code/OSS/cosmos-framework/cosmos_framework/data/vfm/action/datasets/action_sft_dataset.py /Users/hoangvu/Code/OSS/cosmos-framework/tests/action_sft_dataset_test.py
• uvx ruff format --check /Users/hoangvu/Code/OSS/cosmos-framework/cosmos_framework/data/vfm/action/datasets/action_sft_dataset.py /Users/hoangvu/Code/OSS/cosmos-framework/tests/action_sft_dataset_test.py
• PYTHONPATH=/Users/hoangvu/Code/OSS/cosmos-framework uvx pytest --noconftest /Users/hoangvu/Code/OSS/cosmos-framework/tests/action_sft_dataset_test.py -o addopts=