Add _no_split_modules to some models

[a-r-r-o-w]

Adds some of the missing _no_split_modules for internal blocks to Allegro, Hunyuan, CogVideoX and CogView. The tests seem to be passing for the ones modified here. Will look into the remaining in a separate PR - it seems to be happening on models that have a scale_shift_table or similar learned nn.Parameter

CUDA_VISIBLE_DEVICES="2,3" pytest -s tests/models/transformers/ -k test_m
odel_parallelism

FAILED tests/models/transformers/test_models_transformer_aura_flow.py::AuraFlowTransformerTests::test_model_parallelism - RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cuda:1! (when checking argum...
FAILED tests/models/transformers/test_models_transformer_sana.py::SanaTransformerTests::test_model_parallelism - RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:1 and cuda:0!

The number of layers in the test was modified because if we don't, all layers end up on same device and the test fails (because it expects both devices to be used)

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.