docs: simplify f32 casting section — remove migration table

Remove per-task migration status table. The section now explains
the general principle concisely without listing internal task status.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Signed-off-by: Justin Chu <justinchu@microsoft.com>

Author: justinchuby

.agents/skills/multimodal-models/SKILL.md

@@ -279,44 +279,7 @@ Type Error: Type parameter (T) bound to different types
 
 If you're adding a new multimodal model, you don't need to handle this
 manually — mobius inserts the Cast automatically for all encoder graphs.
-
-### Task migration status
-
-Tasks that handle encoder inputs fall into three categories:
-
-**✅ f32 input + Cast pattern (correct):**
-
-| Task | Encoder inputs |
-|------|---------------|
-| `Gemma4Task` | vision `pixel_values`, audio `input_features` |
-| `Phi4mmMultimodalTask` | vision `pixel_values`, audio `audio_embeds` |
-| `VisionLanguage3ModelTask` | vision `pixel_values` |
-| `SpeechLanguageTask` | audio `input_features` |
-| `FunAsrSpeechLanguageTask` | audio `input_features` |
-
-These accept f32 inputs and `Cast(to=config.dtype)` at graph entry.
-
-**⚠️ Needs migration (uses `config.dtype` as input dtype):**
-
-| Task | Input | Issue |
-|------|-------|-------|
-| `VisionLanguageTask` | `pixel_values` | Uses `dtype=config.dtype` — should be f32 + Cast |
-| `MultimodalTask` | `pixel_values`, `audio_features` | Uses `dtype=config.dtype` — should be f32 + Cast |
-| `AudioCTCTask` | `input_features` | Uses `config.dtype or FLOAT` — should always be f32 + Cast |
-
-**✅ Single-model tasks (hardcoded FLOAT — correct):**
-
-| Task | Notes |
-|------|-------|
-| `ImageClassificationTask` | Single-model, f32 input is correct |
-| `AudioFeatureExtractionTask` | Single-model, f32 input is correct |
-| `CodecTask` | Single-model, f32 waveform input |
-| `QwenImageVAETask` | Single-model, f32 input |
-| `ObjectDetectionTask` | Single-model, f32 input |
-
-Single-model tasks don't need the Cast pattern because they're
-standalone ONNX models (not part of a multi-model GenAI pipeline).
-Their f32 inputs are consumed directly.
+If the model dtype is already f32, no Cast is needed.
 
 ## Cross-references