Capture exceptions raised from PyFunctionLibrary functions for debuggability.

[copybara-service]

Capture exceptions raised from PyFunctionLibrary functions for debuggability.

Before this change, when there is an error raised inside a PyFunctionLibrary method, the program would crash with a long stack trace, something like the following:

<super-long c++ stack trace with no meaningful information>

libc++abi: terminating due to uncaught exception of type pybind11::error_already_set: ValueError: Failed to run graph for post-training quantization calibration.

The information in the raised exception is lost in this case, hindering debuggability.

This change instead returns regular value None back to the c++ layer to indicate error, and prints the exception chain from the python side.
It prints out the exact error message one would see when an exception is raised from regular python script.

The return values of PyFunctionLibrary methods are all changed to std::optional<T>, where T is the successful return value and std::nullopt indicates error.
To avoid conflict btw erroneous None, methods that do not essentially return a value now returns a dummy True to indicate successful run.