First-class non-message workflow context / prefetched tool results for `create_agent`

### Checked other resources - [x] This is a feature request, not a bug report or usage question. - [x] I added a clear and descriptive title that summarizes the feature request. - [x] I used the GitHub search to find a similar feature request and didn't find it. - [x] I checked the LangChain documentation and API reference to see if this feature already exists. - [x] This is not related to the langchain-community package. ### Package (Required) - [x] langchain - [ ] langchain-openai - [ ] langchain-anthropic - [ ] langchain-classic - [ ] langchain-core - [ ] langchain-model-profiles - [ ] langchain-tests - [ ] langchain-text-splitters - [ ] langchain-chroma - [ ] langchain-deepseek - [ ] langchain-exa - [ ] langchain-fireworks - [ ] langchain-groq - [ ] langchain-huggingface - [ ] langchain-mistralai - [ ] langchain-nomic - [ ] langchain-ollama - [ ] langchain-openrouter - [ ] langchain-perplexity - [ ] langchain-qdrant - [ ] langchain-xai - [ ] Other / not sure / general ### Feature Description ## Summary `create_agent` should support workflow nodes that compute and inject context **before** the main model step runs. Today, one workaround is to replay precomputed tool output as message history. That worked for us with models until `gemini-2.5-flash-lite`, but it breaks with `gemini-3.1-flash-lite-preview`, which rejects this message pattern. ## Problem Newer Gemini models appear to validate tool-call turn structure more strictly, and reject this pattern. ## Reproduction ```python import pytest from langchain_core.messages import AIMessage, HumanMessage, ToolMessage from langchain_google_genai.chat_models import ChatGoogleGenerativeAI, ChatGoogleGenerativeAIError from app.config import settings pytestmark = [ pytest.mark.integration, ] @pytest.mark.parametrize( ("model_name", "expected_content", "expected_error_match"), [ pytest.param("gemini-2.5-flash-lite", "world", None, id="gemini-2.5-flash-lite"), pytest.param( "gemini-3.…