refactor: improve chat history persistence reliability with metadata-only migration, error handling, and optimized file updates

internal/sse/consumer.go

@@ -36,9 +36,13 @@ func CollectStream(resp *http.Response, thinkingEnabled bool, closeBody bool) Co
 		currentType = "thinking"
 	}
 	_ = deepseek.ScanSSELines(resp, func(line []byte) bool {
-		if chunk, done, parsed := ParseDeepSeekSSELine(line); parsed && !done {
+		chunk, done, parsed := ParseDeepSeekSSELine(line)
+		if parsed && !done {
 			collector.ingestChunk(chunk)
 		}
+		if done {
+			return false
+		}
 		if stopped {
 			return true
 		}
@@ -52,7 +56,8 @@ func CollectStream(resp *http.Response, thinkingEnabled bool, closeBody bool) Co
 				contentFilter = true
 			}
 			// Keep scanning to collect late-arriving citation metadata lines
-			// that can appear after response/status=FINISHED.
+			// that can appear after response/status=FINISHED, but stop as soon
+			// as [DONE] arrives.
 			stopped = true
 			return true
 		}

internal/sse/consumer_edge_test.go

@@ -5,6 +5,7 @@ import (
 	"net/http"
 	"strings"
 	"testing"
+	"time"
 )
 
 // ─── CollectStream edge cases ────────────────────────────────────────
@@ -227,6 +228,39 @@ func TestCollectStreamStatusFinished(t *testing.T) {
 	}
 }
 
+func TestCollectStreamStopsOnDoneAfterFinished(t *testing.T) {
+	pr, pw := io.Pipe()
+	defer func() { _ = pw.Close() }()
+
+	resp := &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     make(http.Header),
+		Body:       pr,
+	}
+
+	resultCh := make(chan CollectResult, 1)
+	go func() {
+		resultCh <- CollectStream(resp, false, false)
+	}()
+
+	_, _ = io.WriteString(pw, "data: {\"p\":\"response/content\",\"v\":\"Hello\"}\n")
+	_, _ = io.WriteString(pw, "data: {\"p\":\"response/status\",\"v\":\"FINISHED\"}\n")
+	_, _ = io.WriteString(pw, "data: {\"p\":\"response/fragments/-1/results\",\"v\":[{\"url\":\"https://example.com/a\",\"cite_index\":1}]}\n")
+	_, _ = io.WriteString(pw, "data: [DONE]\n")
+
+	select {
+	case result := <-resultCh:
+		if result.Text != "Hello" {
+			t.Fatalf("expected text to freeze at FINISHED, got %q", result.Text)
+		}
+		if got := result.CitationLinks[1]; got != "https://example.com/a" {
+			t.Fatalf("expected citation metadata after FINISHED, got %q", got)
+		}
+	case <-time.After(500 * time.Millisecond):
+		t.Fatal("CollectStream did not stop on [DONE] after FINISHED")
+	}
+}
+
 func TestCollectStreamStopsOnContentFilterStatus(t *testing.T) {
 	resp := makeHTTPResponse(
 		"data: {\"p\":\"response/content\",\"v\":\"safe\"}\n" +