The NameError ('false' is not defined) isn't really an error. The GraphUI generates a JSON serialisation of the graph as the "STUDENT_ANSWER" variable. The standard python3 question type attempts to execute the STUDENT_ANSWER as Python code, which it isn't, giving that error. [In Python the boolean literal for "not true" is False, with an uppercase 'f', whereas in JSON it is false, which is why you're getting that particular error.]
I have no immediate explanation of the sandbox errors. If Python is working correctly, then the built-in Graph question types, which are also in Python, should work fine too. SQL not working either is also a mystery; I can't think of any explanation that explains both those problems, so I'd be inclined to put the SQL problem to one side and focus on Python for now.
I think the first thing to do is to edit your failing graph question and click on the Customise and Template debugging checkboxes. Save the question with Validate-on-save unchecked, then preview the question. Look at the generated Python source code to see if there's anything odd in it - perhaps try running it outside of CodeRunner, too. If that doesn't solve the problem, please show me the text in the "Debug: source code from all test runs" section of the output.
Richard