SGLang worker does not expose embeddings.

Nobody has claimed this yet.

Hi,

when running a model with the sglang worker, you get a 404 on the embeddings:

```
curl -X 'POST' \
'https://helmholtz-blablador.fz-juelich.de:8000/v1/embeddings' \
-H 'accept: application/json' \
-H 'Authorization: Bearer MY_API_KEY' \
-H 'Content-Type: application/json' \
-d '{
"model": "Mistral-7B-Instruct-v0.2",
"engine": "string",
"input": "string",
"user": "string",
"encoding_format": "float"
}'
```

The reply is
```
{"object":"error","message":"Not Found","code":50001}%
```

When you call the same model with the slower model_worker, you get the embeddings.

Start by locating the SGLang worker's handling of the /v1/embeddings endpoint and compare it with the slower model_worker, which is reported to expose embeddings. Reproduce the request from the issue and confirm that the SGLang worker no longer returns 404 while the existing model_worker behavior remains available.

Written by the indexing model from the issue text.