fix: volcanic engine embedding endpoint (ep-xxx-xx) not working with multimodal API

[hanliang97]

Problem\n\nVolcanic engine embedding models with `ep-xxx-xx` endpoint IDs fail with "list index out of range" error during credential verification. Only `doubao-xxx` model names worked correctly.\n\n## Root Cause\n\nTwo issues:\n\n1. The code only routed `doubao-embedding-vision-` prefixed models to the multimodal embeddings API (`/embeddings/multimodal`), while `ep-xxx-xx` endpoint models fell through to the standard embeddings API (`/embeddings`) which doesn't support multimodal endpoints.\n\n2. The multimodal embeddings API returns `resp.data` as a single `MultimodalEmbedding` object with an `.embedding` attribute, not a list. The extraction logic was incorrectly treating it as a list.\n\n## Fix\n\n- Unify all volcanic engine embedding calls to use `multimodal_embeddings.create()` API (compatible with both pure text and multimodal inputs)\n- Fix `_extract_embedding` to check `hasattr(data, 'embedding')` first, correctly handling single MultimodalEmbedding objects\n- Both `ep-xxx-xx` endpoint IDs and `doubao-xxx` model names now work correctly\n\n## Testing\n\nTested with endpoint `ep-20260527164729-5mshx` (doubao-embedding-vision based) - credential verification and embedding generation both pass.

[hanliang97]

为了解决 火山向量模型 sp-xxx-xxx无法添加问题