Persist Ollama token usage from streamed LLM responses (#16) (#38)

## Summary
- Fixes token tracking for [#16](#16): streaming chat never persisted `PromptMetric.tokens_in` / `tokens_out` (admin + account usage showed `—`).
- Drop `StrOutputParser` on async LLM/RAG/data-analysis chains so Ollama `generation_info` (`prompt_eval_count` / `eval_count`) survives; collect usage while streaming via `TokenUsageCollector`.
- Stop calling `self.close()` in `disconnect` (fixes Grafana `Unexpected ASGI message 'websocket.close'`).

## Test plan
- [x] Unit tests: `test_utils`, consumers, LLM/RAG/data-analysis services, finance quotas
- [ ] Deploy / local: send a chat prompt, confirm admin Prompt Metrics shows Tokens In/Out
- [ ] Reload Account usage card — in/out no longer `—` for new turns
- [ ] Confirm WS disconnect no longer raises double-close in logsReviewed-on: #38

llm_be/chat_backend/consumers_graph.py

@@ -23,6 +23,8 @@ from .services.moderation_classifier import moderation_classifier, ModerationLab
 from .services.prompt_classifier.prompt_classifier import PromptClassifier, PromptType
 from .services.data_analysis_service import AsyncDataAnalysisService
 from .utils import (
+    TokenUsageCollector,
+    aiter_text_chunks,
     extract_token_usage,
     has_usable_user_prompt,
     is_heartbeat_payload,
@@ -323,7 +325,9 @@ class ChatConsumerGraph(AsyncWebsocketConsumer):
         await self.accept()
 
     async def disconnect(self, close_code):
-        await self.close()
+        # Connection already closing — do not call self.close() again
+        # (triggers ASGI 'websocket.close' after close completed).
+        pass
 
     async def send_json_message(self, data_str):
         try:
@@ -453,24 +457,29 @@ class ChatConsumerGraph(AsyncWebsocketConsumer):
                 await self.send("START_OF_THE_STREAM_ENDER_GAME_42")
 
                 full_response = ""
-                
+                tokens_in = tokens_out = None
+
                 if isinstance(response_generator_or_dict, dict):
                     content = response_generator_or_dict.get("content", "")
                     await self.send_json_message(json.dumps(response_generator_or_dict))
                     full_response = content
+                    tokens_in, tokens_out = extract_token_usage(
+                        response_generator_or_dict
+                    )
                 else:
-                    async for chunk in response_generator_or_dict:
+                    # Stream raw LLM chunks so final Ollama generation_info
+                    # (prompt_eval_count / eval_count) is not stripped.
+                    usage = TokenUsageCollector()
+                    async for chunk in aiter_text_chunks(
+                        response_generator_or_dict, usage
+                    ):
                         full_response += chunk
                         await self.send_json_message(chunk)
+                    tokens_in, tokens_out = usage.pair
 
                 await self.send("END_OF_THE_STREAM_ENDER_GAME_42")
                 
                 await save_generated_message(conversation_id, full_response)
-                tokens_in, tokens_out = extract_token_usage(
-                    response_generator_or_dict
-                    if isinstance(response_generator_or_dict, dict)
-                    else None
-                )
                 await finish_prompt_metric(
                     prompt_metric,
                     len(full_response),