Enhancement to `mlc.dataclasses.c_class`

[potatomashed]

c_class is used to reflect a class defined on C++ side to Python in zero copy, without the overhead of pybind or any other 3rdparty FFI tools.

However, the features supported in c_class is not on parity with py_class yet. Namely:

• The default __str__ method is not as good looking as py_class's feat(dataclasses): Introduce mlc.dataclasses.stringify #69
• frozen=True is not supported
• Per-field init=False is not supported yet.

Those are easy features to implement but lacking. Contributions are welcome!

[junrushao-openai]

Regarding __str__: would be nice to have a method under mlc.dataclasses. that prints like python dataclass format, plus their id