Add diffusers model_index.json and scheduler config for DFlashPipeline.from_pretrained

This adds the two files needed for `diffusers` to load the pipeline directly via `DFlashPipeline.from_pretrained`:

- `model_index.json` — declares `DFlashPipeline` as the pipeline class and points
`_target_model_name_or_path` to the target/verifier model
- `scheduler/scheduler_config.json` — config for `DFlashTokenDiffusionScheduler`

After merging, the pipeline can be loaded with:

```python
import torch
from diffusers import DFlashPipeline

pipe = DFlashPipeline.from_pretrained(
"z-lab/Qwen3-4B-DFlash-b16",
torch_dtype=torch.bfloat16,
device_map="auto",
)
out = pipe(prompt="Hello, world!")
print(out.texts[0])
```

model_index.json

@@ -0,0 +1,9 @@
+{
+  "_class_name": "DFlashPipeline",
+  "_diffusers_version": "0.40.0.dev0",
+  "_target_model_name_or_path": "Qwen/Qwen3-4B",
+  "scheduler": [
+    "diffusers",
+    "DFlashTokenDiffusionScheduler"
+  ]
+}

scheduler/scheduler_config.json

@@ -0,0 +1,4 @@
+{
+  "_class_name": "DFlashTokenDiffusionScheduler",
+  "_diffusers_version": "0.40.0.dev0"
+}