Refactor stub generator

[alfonsoSR]

Rationale

The goal of this PR is to replace mypy.stubgen and pybind11-stubgen with custom stub generators for Python and C++ extensions.

Bad features of current stub generator

• pybind11-stubgen is not particularly well maintained, it does not have documentation, it is not good at processing enumerations, and it produces rough stubs that require significant post-processing.
• mypy.stubgen is very slow, and the stubs it produces also require a lot of post-processing

Nice features of the new stub generators

• We can discourage users from using deprecated functions/classes/modules by not generating stubs for them. This means that deprecated code still works, but it no longer gets syntax highlighting, so new people will not find it.
• We can raise custom warnings to help us find code that does not follow some rule. For example, we can raise a warning when a function does not have documentation, or when it has C++ types in its signature.
• We can use a logger to control what gets printed to the terminal while generating the stubs. For example, we can use log.warning for warnings that matter to users, and log.debug for information that is only relevant to developers, and then decide to only print developer information if we call the build.py script with the --debug flag.
• Instead of having to define post-processing functions to fix the stubs produced by pybind11-stubgen, we can adapt the stub generators to our needs and simplify the implementation.

TODO

• Fix exposure of enumerations
• Try to fix representation of default arguments: Error rendering docstrings #202
• Incorporate forgotten fix from tudat-bundle: Allow for nested build directories tudat-bundle#40