Provide a way to use a custom base class instead of betterproto.Message

### Summary

Allow to customize the base class, to provide helper methods and enrich the model

### What is the feature request for?

The core library

### The Problem

We developed an internal library on top of betterproto and pika rabbitmq packages.
During development, we found that

1. betterproto lacks a way to customize the base class for the generated messages classes.
2. it's not possible to access message classes in the following way:

```python
import mylib as ml

ml.messages.device.settings.ConfigureDevice(*args)
```

### The Ideal Solution

To ease all above, would be handy to provide:

1. A way to use a `MyMessageBaseClass(betterproto.Message)` as base class for generated messages. This would probably be a new cli argument for the compiler e.g. `--base_class=mylib.base.MyMessageBaseClass` to use in the init jinja template.
2. import subpackages (if any) in the `__init__.py` generated files.

### The Current Solution

To enrich the classes with helper methods and to override some methods of the betterproto.Message class, we first patched the generated classes, to add mixins and overridden methods at runtime.

This is working well, but Python IDEs' support is limited.

We solved some of the issues above by adding a post compilation step at installation time, to add code to the betterproto generated files.
During this step, we also add subpackages imports to `__init__.py` files, to access messages conveniently using the fqn (see point 2. of the issue).