[Bug]: Backend server crashes silently on startup

### File Name deep-search-power ### What happened? Title: Backend server crashes silently on startup Description: I am unable to run the deep-search agent. The backend server crashes instantly and silently upon launch, which causes the frontend to show a "Waiting for backend..." message and produce a continuous stream of Error: connect ECONNREFUSED 127.0.0.1:8000 errors. There are no tracebacks or error messages produced by the backend process when it crashes. Environment: (Please describe your environment here, e.g., "Google Cloud IDE," "local VS Code dev container," etc.) Project version: deep-search (formerly gemini-fullstack) Debugging Steps Taken: I have worked extensively with an AI assistant to debug this issue. We have taken the following steps without success: Set Environment Variables: Confirmed that GEMINI_API_KEY is correctly set in the environment. Fixed ImportError: The first crashes were caused by ImportError: cannot import name 'Tool' from 'google.adk.tools'. We resolved this by modifying my-fullstack-agent/pyproject.toml to pin the dependency to google-adk==1.8.0 and running uv pip sync. Handled Missing Search Credentials: We identified that my-fullstack-agent/app/tools.py would raise a ValueError if GOOGLE_API_KEY and GOOGLE_CSE_ID were not set. We modified the _google_search function to return an error message instead of crashing, but this did not solve the silent crash. Attempted Direct Server Launch: We have tried to bypass the Makefile scripts to get a clearer error message. Running uv run uvicorn app.main:app --host 0.0.0.0 --port 8000 from the my-fullstack-agent directory also results in an instant, silent crash with no output. Attempted Playground Mode: Running make playground also fails to start a persistent server process. Confirmed Working Directory: Ensured all make commands are run from the correct my-fullstack-agent subdirectory. Current Status: The backend is completely non-functional. The silent nature of the crash, even after fixing several definite bugs, suggests a deeper, low-level issue within the application's startup sequence. We are unable to gather any more diagnostic information. ### Relevant log output ```shell ``` ### Code of Conduct - [x] I agree to follow this project's Code of Conduct