Add from_single_file support for Ideogram4Transformer2DModel

[HaozheZhang6]

What does this PR do?

Adds from_single_file support for Ideogram4Transformer2DModel, so the community Ideogram 4 single-file / GGUF checkpoints load directly — e.g. leejet/ideogram-4-GGUF and stduhpf/ideogram-4-gguf, which are popular but currently can't be loaded in diffusers.

The single-file layout already matches diffusers except attention, which fuses the projection as attention.qkv and names the output projection attention.o. The conversion splits qkv into to_q/to_k/to_v and renames o → to_out.0; every other key passes through unchanged.

import torch
from diffusers import Ideogram4Transformer2DModel, GGUFQuantizationConfig

transformer = Ideogram4Transformer2DModel.from_single_file(
    "https://huggingface.co/leejet/ideogram-4-GGUF/blob/main/ideogram4-Q4_0.gguf",
    config="ideogram-ai/ideogram-4-fp8",
    subfolder="transformer",
    quantization_config=GGUFQuantizationConfig(compute_dtype=torch.bfloat16),
)

Checked on the Q4_0 GGUF: the converted state dict matches the model 1:1 (all keys and shapes), it loads through from_single_file and runs on a 16 GB GPU (~5.7 GB), and the split to_q/to_k/to_v line up with the official bf16 weights — cross-checked against the swapped splits to confirm the [q, k, v] order.

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?
• Was this discussed/approved via a GitHub issue or the forum? It's a small additive from_single_file registration, mirroring the existing converters (flux/chroma/cosmos/…).
• Did you make sure to update the documentation with your changes? No doc change needed.
• Did you write any new necessary tests? The Ideogram weights are gated, so a standard single-file integration test needs gated access in CI — happy to add one in whatever form you prefer; verification is in the description above.

Who can review?

@DN6 @sayakpaul

[HaozheZhang6]

Folded this into #13942 since both add single_file support to the same two files. Easier to review as one. Closing here.