Ensure all model response chunks are streamed without filtering

Remove logic that filtered out streaming chunks, ensuring all content, including thinking tokens, is passed through to the client.

Replit-Commit-Author: Agent
Replit-Commit-Session-Id: a225ab96-d6aa-42ce-a50b-aa844bc8db2b
Replit-Commit-Checkpoint-Type: full_checkpoint
Replit-Commit-Event-Id: d0de45b0-d639-4150-bca9-9b990b65bda9
Replit-Commit-Screenshot-Url: https://storage.googleapis.com/screenshot-production-us-central1/cbd1ae84-d682-4025-b26d-c56df1ac13a1/a225ab96-d6aa-42ce-a50b-aa844bc8db2b/jm4hNZf
Replit-Helium-Checkpoint-Created: true

main.go

@@ -209,7 +209,6 @@ func streamUpstream(w http.ResponseWriter, modelID string, messages []Message, t
         flusher, _ := w.(http.Flusher)
         scanner := bufio.NewScanner(resp.Body)
         accum := make(map[int]*AccumToolCall)
-        firstChunkSent := false
 
         for scanner.Scan() {
                 line := scanner.Text()
@@ -296,15 +295,6 @@ func streamUpstream(w http.ResponseWriter, modelID string, messages []Message, t
                         continue
                 }
 
-                if !firstChunkSent && delta["content"] != nil {
-                        delta["role"] = "assistant"
-                        firstChunkSent = true
-                }
-
-                if delta["content"] == nil && delta["tool_calls"] == nil && finishReason == nil {
-                        continue
-                }
-
                 // podmień model na alias klienta
                 chunk["model"] = clientModel
                 out, _ := json.Marshal(chunk)