Modular: three tiny test fixtures cannot serve the new workflow tests

> Opened by Claude (Opus 5) on behalf of @yiyixuxu.

### Describe the bug

#14355 adds three shared modular tests — `test_from_pretrained_workflow`, `test_load_components_workflow` and `test_unload_components` — that build a pipeline from the test class's `pretrained_model_name_or_path` and compare it against the blocks the test class declares:

```python
pipe = ModularPipeline.from_pretrained(self.pretrained_model_name_or_path, workflow=workflow_name)
ref_pipe = blocks.get_workflow(workflow_name).init_pipeline(self.pretrained_model_name_or_path)
assert set(pipe.component_names) == set(ref_pipe.component_names)
```

Three `hf-internal-testing` fixtures cannot answer that, in three different ways. None of them is a problem with the pipelines' code — they are the first tests to go through `from_pretrained` for these classes at all, since `ModularPipelineTesterMixin.get_pipeline` builds from `self.pipeline_blocks_class()` directly. They are skipped with a TODO each on that branch; this issue tracks the fixtures.

### 1. `hf-internal-testing/tiny-anima-modular-pipe` does not carry a `modular_model_index.json`

```
OSError: Failed to load config from 'hf-internal-testing/tiny-anima-modular-pipe'.
Could not find or load 'modular_model_index.json' or 'model_index.json'.
```

Anima's test classes override `get_pipeline` to assemble dummy components in-process:

```python
def get_pipeline(self, components_manager=None, dtype=torch.float32):
pipe = self.pipeline_blocks_class().init_pipeline(components_manager=components_manager)
pipe.update_components(**get_dummy_components())
```

so `pretrained_model_name_or_path` was never loaded from and the repository was never needed. **Fix:** publish a tiny Anima modular repository under that name. Affects `TestAnimaModularPipelineFast` and `TestAnimaImg2ImgModularPipelineFast`, three tests each.

### 2. `hf-internal-testing/tiny-flux2-klein-modular` names the base blocks for a distilled checkpoint

```json
{
"_blocks_class_name": "Flux2KleinBaseAutoBlocks",
"_class_name": "Flux2KleinModularPipeline",
"is_distilled": true
}
```

`from_pretrained` honours `_blocks_class_name`, so it builds the **base** blocks while the test classes declare `Flux2KleinAutoBlocks`. The two differ by exactly the guider:

```
Flux2KleinAutoBlocks text2image ['image_processor', 'scheduler', 'text_encoder', 'tokenizer', 'transformer', 'vae']
Flux2KleinBaseAutoBlocks text2image ['guider', ...same...]
```

The route reads identically by block name on both sides, which is why only the component set shows it; the classes differ at `text_encoder` (`Flux2KleinBaseTextEncoderStep`) and `denoise.denoise.denoiser` (`Flux2KleinBaseLoopDenoiser`).

**Fix:** set `_blocks_class_name` to `Flux2KleinAutoBlocks`, since the other two fields both say distilled — or, if the fixture is meant to be the base model, set `_class_name` to `Flux2KleinBaseModularPipeline`, `is_distilled` to `false`, and have the test classes declare `Flux2KleinBaseAutoBlocks`. Affects `TestFlux2KleinModularPipelineFast` and `TestFlux2KleinImageConditionedModularPipelineFast`.

### 3. `hf-internal-testing/tiny-qwenimage-edit-modular` names the base pipeline class

```json
{
"_blocks_class_name": "SequentialPipelineBlocks",
"_class_name": "QwenImageModularPipeline"
}
```

`_blocks_class_name` is a saved generic sequence, so resolution falls back to `_class_name`'s `default_blocks_name`, i.e. `QwenImageAutoBlocks`:

```
ValueError: Workflow 'image_conditioned' not found in QwenImageAutoBlocks.
Available workflows: ['text2image', 'image2image', 'inpainting', 'controlnet_text2image',
'controlnet_image2image', 'controlnet_inpainting']
```

**Fix:** record the Edit classes. `hf-internal-testing/tiny-qwenimage-edit-plus-modular` carries the same `_class_name: QwenImageModularPipeline` and is worth correcting at the same time — it does not fail today only because `QwenImageEditPlusAutoBlocks` declares no workflow map, so the test skips itself. Affects `TestQwenImageEditModularPipelineFast`.

### Reproduction

On the branch of #14355:

```
pytest tests/modular_pipelines/ -k "test_from_pretrained_workflow or test_load_components_workflow or test_unload_components"
```

Before the skips: 10 failed, 79 passed.