Fix SanaTransformer2DModel crash when guidance_embeds=True with non-guidance pipelines

[Ricardo-M-L]

What does this PR do?

Fixes #12540.

When `SanaTransformer2DModel` is built with `guidance_embeds=True` and then driven by a pipeline that does not thread `guidance` through (`SanaPipeline`, `SanaPAGPipeline`), the transformer forward crashes with:

```
TypeError: SanaCombinedTimestepGuidanceEmbeddings.forward() got an unexpected keyword argument 'batch_size'
```

Root cause

`SanaTransformer2DModel.forward` routed the time-embedding call on whether `guidance` was passed:

```python
if guidance is not None:
timestep, embedded_timestep = self.time_embed(timestep, guidance=guidance, hidden_dtype=...)
else:
timestep, embedded_timestep = self.time_embed(timestep, batch_size=batch_size, hidden_dtype=...)
```

But the two embedding classes have incompatible signatures:

• `AdaLayerNormSingle.forward` takes `batch_size` and does not accept `guidance`
• `SanaCombinedTimestepGuidanceEmbeddings.forward` takes `guidance` and does not accept `batch_size`

So with `guidance_embeds=True` and no guidance kwarg, the `batch_size` branch is hit against the wrong embedder and blows up.

Fix

Dispatch on the embedding type instead of on the input. If the model was configured with guidance embeddings but the caller omitted `guidance`, raise a clear `ValueError` naming the misconfiguration rather than surfacing a cryptic kwargs error from deeper in the embedder.

Reproducer (from #12540)

```python
from diffusers import SanaTransformer2DModel, SanaPAGPipeline
import torch

config = SanaTransformer2DModel.load_config('Efficient-Large-Model/Sana_600M_512px_diffusers', subfolder='transformer')
config['guidance_embeds'] = True
new_transformer = SanaTransformer2DModel.from_config(config)

pipe = SanaPAGPipeline.from_pretrained(
'Efficient-Large-Model/Sana_1600M_512px_MultiLing_diffusers',
pag_applied_layers=['transformer_blocks.8'],
transformer=new_transformer,
torch_dtype=torch.bfloat16,
vae=None,
).to('cuda')
pipe(prompt='cat', output_type='latent', height=768, width=512, pag_scale=1.0, guidance_scale=4.0, num_inference_steps=20)
```

Before: `TypeError: SanaCombinedTimestepGuidanceEmbeddings.forward() got an unexpected keyword argument 'batch_size'`
After: `ValueError: SanaTransformer2DModel was configured with guidance_embeds=True, but guidance was not provided. Either pass guidance to the transformer or rebuild the model with guidance_embeds=False.`

Testing

Manually built tiny SanaTransformer2DModel instances and confirmed:

1. `guidance_embeds=True` without `guidance` → clear ValueError
2. `guidance_embeds=True` with `guidance` → works
3. `guidance_embeds=False` (default, AdaLayerNormSingle) → unchanged, works

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?
• Did you read our philosophy doc (important for complex PRs)?
• 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?

@yiyixuxu @DN6 @dg845