fix: Resolve socket hang up errors for large file execution

[usnavy13]

Summary

• Fixes "socket hang up" errors when executing code with large files (>40MB) from LibreChat
• Root cause: Node.js 20 sets a default 5-second socket timeout; long-running server operations (cold sandbox starts, large file mounting) exceed this, killing the TCP connection
• Three complementary server-side fixes applied without any client code changes

Changes

Streaming keepalive on /exec (src/api/exec.py)

• Converts the endpoint to use StreamingResponse that sends a space character every 3 seconds while execution is running
• JSON parsers ignore leading whitespace, so this is fully transparent to all clients
• Properly re-raises ValidationError/ServiceUnavailableError so FastAPI exception handlers still return correct HTTP status codes
• Preserves OpenAPI schema via responses parameter

Non-blocking file I/O (src/services/file.py, src/services/execution/runner.py)

• Fixed get_file_content(): response.read() was running synchronously on the asyncio event loop, blocking all concurrent HTTP connections during large file downloads from MinIO
• Added stream_file_to_path() using MinIO's fget_object for direct disk-to-disk transfer without loading files into memory
• Updated _mount_files_to_sandbox() to use streaming, with non-blocking os.chown/os.chmod
• Added stream_file_to_path to FileServiceInterface

Configuration (.env)

• Increased SANDBOX_POOL_PY from 2 to 5 to reduce cold-start frequency

Test plan

• New functional test: tests/functional/test_concurrent_file_exec.py — uploads 50MB CSV, fires 5 concurrent file execs + 3 simple pings, asserts pings are not blocked
• Verified fix from LibreChat container using node-fetch (exact CodeExecutor flow)
• Verified existing functional tests pass
• Lint checks pass (black, flake8)

🤖 Generated with Claude Code