/**
 * Append a decoded SSE chunk to the running buffer and extract complete data lines.
 *
 * Returns the raw JSON strings from each `data: {...}` line and the updated
 * buffer (partial-line remainder). Skips the `[DONE]` sentinel.
 *
 * Both the vLLM server-side reader and the client-side chat reader share
 * this logic — keeping buffer management in one place prevents drift when
 * the SSE wire format changes.
 */
export function consumeSSEChunk(
  buffer: string,
  chunk: string
): { jsonLines: string[]; buffer: string } {
  const full = buffer + chunk;
  const parts = full.split("\n");
  const newBuffer = parts.pop()!;
  const jsonLines: string[] = [];

  for (const part of parts) {
    const trimmed = part.trim();
    if (!trimmed.startsWith("data: ") || trimmed === "data: [DONE]") continue;
    jsonLines.push(trimmed.slice(6));
  }

  return { jsonLines, buffer: newBuffer };
}