docs(inference): fix unstable import path in embeddings how-to

## Purpose

The embeddings how-to showed importing EmbedModel via the internal
`pinecone.client.inference` module path, which is not part of the
stable public API and is subject to change without notice.

## Solution

Replace `from pinecone.client.inference import Inference` /
`Inference.EmbedModel.Multilingual_E5_Large` with the canonical
`from pinecone import Pinecone, EmbedModel` / `EmbedModel.Multilingual_E5_Large`,
which is the stable public export guaranteed by `pinecone.__all__`.

Author: jhamon

docs/how-to/inference/embeddings.md

@@ -69,13 +69,12 @@ Use the {class}`~pinecone.models.enums.EmbedModel` enum for tab-completion and t
 safety:
 
 ```python
-from pinecone import Pinecone
-from pinecone.client.inference import Inference
+from pinecone import Pinecone, EmbedModel
 
 pc = Pinecone(api_key="your-api-key")
 
 result = pc.inference.embed(
-    model=Inference.EmbedModel.Multilingual_E5_Large,
+    model=EmbedModel.Multilingual_E5_Large,
     inputs=["search query"],
     parameters={"input_type": "query"},
 )