refactor: replace processed output comparison with baseline-based validation in SSE simulator

internal/sse/content_filter_leak.go

@@ -8,28 +8,32 @@ func filterLeakedContentFilterParts(parts []ContentPart) []ContentPart {
 	}
 	out := make([]ContentPart, 0, len(parts))
 	for _, p := range parts {
-		cleaned := stripLeakedContentFilterSuffix(p.Text)
-		if shouldDropCleanedLeakedChunk(cleaned) {
+		cleaned, stripped := stripLeakedContentFilterSuffix(p.Text)
+		// Only drop the chunk when we actually stripped a leaked CONTENT_FILTER
+		// suffix. Plain whitespace chunks are valid SSE content and must stay.
+		if stripped && shouldDropCleanedLeakedChunk(cleaned) {
 			continue
 		}
-		p.Text = cleaned
+		if stripped {
+			p.Text = cleaned
+		}
 		out = append(out, p)
 	}
 	return out
 }
 
-func stripLeakedContentFilterSuffix(text string) string {
+func stripLeakedContentFilterSuffix(text string) (string, bool) {
 	if text == "" {
-		return text
+		return text, false
 	}
 	upperText := strings.ToUpper(text)
 	idx := strings.Index(upperText, "CONTENT_FILTER")
 	if idx < 0 {
-		return text
+		return text, false
 	}
 	// Keep "\n" so we don't collapse line structure when the upstream model
 	// appends leaked CONTENT_FILTER markers after a line break.
-	return strings.TrimRight(text[:idx], " \t\r")
+	return strings.TrimRight(text[:idx], " \t\r"), true
 }
 
 func shouldDropCleanedLeakedChunk(cleaned string) bool {

internal/sse/line_test.go

@@ -63,6 +63,16 @@ func TestParseDeepSeekContentLineContent(t *testing.T) {
 	}
 }
 
+func TestParseDeepSeekContentLinePreservesSpaceOnlyChunk(t *testing.T) {
+	res := ParseDeepSeekContentLine([]byte(`data: {"v":" "}`), false, "text")
+	if !res.Parsed || res.Stop {
+		t.Fatalf("expected parsed non-stop result: %#v", res)
+	}
+	if len(res.Parts) != 1 || res.Parts[0].Text != " " || res.Parts[0].Type != "text" {
+		t.Fatalf("unexpected parts for space-only chunk: %#v", res.Parts)
+	}
+}
+
 func TestParseDeepSeekContentLineStripsLeakedContentFilterSuffix(t *testing.T) {
 	res := ParseDeepSeekContentLine([]byte(`data: {"p":"response/content","v":"正常输出CONTENT_FILTER你好，这个问题我暂时无法回答"}`), false, "text")
 	if !res.Parsed || res.Stop {