Pass binary payloads to `decode_request` unchanged

Nobody has claimed this yet.

## 🚀 Feature

Pass binary payloads to `decode_request` unchanged.

### Motivation

We use LitServe for BYOC SageMaker Inference endpoints that take protobuf requests of the following format:

```http
POST /invocations
Content-Type: application/x-recordio-protobuf

```

LitServe currently parses all non-form/multipart request requests as JSON. To support `application/x-recordio-protobuf`, we override private methods:

```python

class _RawRequestHandler(RegularRequestHandler):
"""Read the raw protobuf body instead of parsing JSON. The returned dict (bytes + str)
is picklable, so it crosses the worker transport cleanly."""

async def _prepare_request(
self, request: Request, request_type: Any
) -> dict[str, Any]:
return {
"body": await request.body(),
"content_type": request.headers.get("Content-Type", ""),
}

class RankerLitServer(ls.LitServer):
def _register_api_endpoints(
self, lit_api: ls.LitAPI, request_type: Any, response_type: Any
) -> None:
handler = _RawRequestHandler(lit_api, self)

async def endpoint_handler(request: Request) -> Response:
return await handler.handle_request(request, Request)

self.app.add_api_route(
lit_api.api_path,
endpoint_handler,
methods=["POST"],
dependencies=[Depends(self.setup_auth())],
)
```

While it works, overriding private APIs makes future LitServe upgrades brittle.

### Pitch

Enable LitServe to pass bytes from the request body to `decode_request`. This can either be opt-in manually, or via introspecting the `Content-type` header. For raw byte bodies, LitServe
should:

1. Read the request via `await request.body()` without JSON parsing.
2. Pass the resulting `bytes` to `decode_request`.
3. Make the request content type available to the API.
4. Preserve the existing worker-transport and batching behavior.
5. Leave existing JSON and multipart handling unchanged by default.

### Alternatives

- **Private handler override:** This is our current solution, but it relies on
internal methods that may change between LitServe releases.

### Additional context

Past issue #362

Start by tracing the request path through _register_api_endpoints, _prepare_request, handle_request, and decode_request, including the existing JSON and multipart handling. Check how worker transport and batching are preserved, then verify that binary request bodies and their Content-Type reach the API unchanged while existing defaults remain unchanged.

Written by the indexing model from the issue text.