[Oaklight]

Why

httpserver can already stream a response, but nothing in the project turns
application data into a valid text/event-stream body. Every server-side SSE
user therefore hand-rolls the same formatting rules — and gets the edge cases
wrong: a data value containing \n silently truncates on clients that
dispatch per blank line, an id with a newline corrupts the stream, and a
negative retry is accepted by nobody.

This adds the producer half of the protocol next to the existing consumer, in
the same single file, so the two sides are verified against each other.

What

• encode_event(data="", *, event=None, id=None, retry=None) returns one
  complete block, including the terminating blank line that dispatches it.
  Multi-line payloads produce one data: field per line; CRLF and CR are
  normalized. A line break in event/id, NUL in id, or a retry that is
  not a non-negative integer raises ValueError.
• encode_comment(text="") returns a comment block for heartbeats.
• Both are pure functions with no network dependency, so they work with
  httpserver.StreamingResponse, any other transport, or a direct write().

The helpers sit in the same module as EventSource, which makes a round-trip
test possible: whatever encode_event emits, the parser must reconstruct.

Docs

• docs/modules/sse.md (EN + ZH): title, feature list, a server-side encoding
  example, API reference entries, comparison table.
• Module index descriptions and changelog entries in both languages.

Tests

12 new cases in TestEncodeEvent cover single-line and multi-line payloads,
CRLF/CR normalization, every optional field, empty data, trailing newlines,
encoder → parser round-trip, and each rejection path. pytest sse/test_sse_correctness.py passes
(71 tests).

Fixes #241