2.71.1
(fix): Fix SSE stream reconnection correctness to match the reference implementation:
(1) Do not reconnect when no terminator is configured (the client cannot
distinguish a completed stream from a dropped connection).
(2) Apply a default 1-second backoff between reconnect attempts when the
server sends no retry: directive (previously zero delay).
(3) Use the last dispatched event ID for reconnection rather than the
last parsed id, preventing silent event loss on mid-event drops.
(4) Treat reconnect function failures as consumed attempts; throw
IOException with the last failure as InnerException when the cap is
exhausted, instead of silently truncating the stream.
(5) Let OperationCanceledException propagate immediately from reconnectFn
instead of swallowing it as a retry attempt.
(6) Dispose the old HTTP response only after a new one is successfully
obtained, preventing ObjectDisposedException on retry.