[ML] Server-Sent Events for Inference response

Initial implementation of streaming inference responses as Server-Sent
Events for the `POST /_inference` API.

Bytes are requested and read from a Flow.Publisher and encoded in a
ChunkedRestResponseBodyPart before sent to the REST channel.  The
channel will request more bytes via the `getNextPart` API.

Encoding is done in two parts:
1. A wrapper encoding to format the messages as a Server-Sent Event
   stream.
2. The existing JSON (or requested) encoding for the data payload using
   XContent.

Example messages:
```
event: message
data: { "completion": [{"delta": "hello, world"}] }

```