Fix: too long context during KG issue. (#5723)

### What problem does this PR solve?

#5088

### Type of change

- [x] Bug Fix (non-breaking change which fixes an issue)

api/db/services/document_service.py

 from api.db.db_utils import bulk_insert_into_db
 from api import settings
 from api.utils import current_timestamp, get_format_time, get_uuid
-from graphrag.general.mind_map_extractor import MindMapExtractor
 from rag.settings import SVR_QUEUE_NAME
 from rag.utils.storage_factory import STORAGE_IMPL
 from rag.nlp import search, rag_tokenizer
         cks = [c for c in docs if c["doc_id"] == doc_id]
 
         if parser_ids[doc_id] != ParserType.PICTURE.value:
+            from graphrag.general.mind_map_extractor import MindMapExtractor
             mindmap = MindMapExtractor(llm_bdl)
             try:
                 mind_map = trio.run(mindmap, [c["content_with_weight"] for c in docs if c["doc_id"] == doc_id])

graphrag/general/extractor.py

 from graphrag.utils import get_llm_cache, set_llm_cache, handle_single_entity_extraction, \
     handle_single_relationship_extraction, split_string_by_multi_markers, flat_uniq_list, chat_limiter
 from rag.llm.chat_model import Base as CompletionLLM
+from rag.prompts import message_fit_in
 from rag.utils import truncate
 
 GRAPH_FIELD_SEP = "<SEP>"
         response = get_llm_cache(self._llm.llm_name, system, hist, conf)
         if response:
             return response
-        response = self._llm.chat(system, hist, conf)
+        _, system_msg = message_fit_in([{"role": "system", "content": system}], int(self._llm.max_length * 0.97))
+        response = self._llm.chat(system_msg[0]["content"], hist, conf)
         response = re.sub(r"<think>.*</think>", "", response, flags=re.DOTALL)
         if response.find("**ERROR**") >= 0:
             raise Exception(response)