fix: Bump max_tokens to 65536 (model supports 94K context, extraction needs headroom)

.env.example

@@ -30,7 +30,7 @@ LLM_FALLBACK_MODEL=fyn-qwen35-chat
 #LLM_STAGE5_MODALITY=thinking
 
 # Max tokens for LLM responses (OpenWebUI defaults to 1000 — pipeline needs much more)
-LLM_MAX_TOKENS=32768
+LLM_MAX_TOKENS=65536
 
 # Embedding endpoint (Ollama container in the compose stack)
 EMBEDDING_API_URL=http://chrysopedia-ollama:11434/v1

backend/config.py

@@ -44,7 +44,7 @@ class Settings(BaseSettings):
     llm_stage5_modality: str = "chat"
 
     # Max tokens for LLM responses (OpenWebUI defaults to 1000 which truncates pipeline JSON)
-    llm_max_tokens: int = 32768
+    llm_max_tokens: int = 65536
 
     # Embedding endpoint
     embedding_api_url: str = "http://localhost:11434/v1"