Expose the correct namespaces for Python modules

[datapythonista]

Currently, in the datafusion module there is no match between what dir() returns, and what's actually inside the module. For example:

>>> import datafusion

>>> dir(datafusion)
['ABCMeta', 'Accumulator', 'AggregateUDF', 'Alias', 'Analyze', 'Between', 'Case', 'Cast', 'Config', 'CreateMemoryTable', 'CreateView', 'DFSchema', 'DataFrame', 'Distinct', 'DropTable', 'Exists', 'Explain', 'Expr', 'Extension', 'Filter', 'GroupingSet', 'ILike', 'InList', 'InSubquery', 'IsFalse', 'IsNotFalse', 'IsNotNull', 'IsNotTrue', 'IsNotUnknown', 'IsTrue', 'IsUnknown', 'Like', 'Limit', 'List', 'Negative', 'Not', 'Partitioning', 'Placeholder', 'Projection', 'Repartition', 'RuntimeConfig', 'SQLOptions', 'ScalarSubquery', 'ScalarUDF', 'ScalarVariable', 'SessionConfig', 'SessionContext', 'SimilarTo', 'Sort', 'Subquery', 'SubqueryAlias', 'TableScan', 'TryCast', 'Window', 'WindowFrame', '__all__', '__builtins__', '__cached__', '__doc__', '__file__', '__loader__', '__name__', '__package__', '__path__', '__spec__', '__version__', '_internal', 'abstractmethod', 'col', 'column', 'common', 'expr', 'importlib_metadata', 'lit', 'literal', 'pa', 'udaf', 'udf']

>>> datafusion.ABCMeta  # <- THIS IS A STANDARD LIBRARY CLASS, NOT PART OF DATAFUSION
<class 'abc.ABCMeta'>

>>> dir(datafusion.common)  # <- BASED ON THIS, THERE IS NOTHING IN THE MODULE EXCEPT THE MODULE ITSELF AGAIN
['__builtins__', '__cached__', '__doc__', '__file__', '__getattr__', '__loader__', '__name__', '__package__', '__spec__', 'common']

>>> datafusion.common.common
<module 'common'>

>>> datafusion.common.SqlTable  # <- BUT WHEN USING THE MODULE THERE ARE ACTUALLY CLASSES
<class 'datafusion.common.SqlTable'>

With this PR, I basically show what is part of the module, and I hide what's being displayed now as a side effect:

>>> import datafusion

>>> dir(datafusion)
['Accumulator', 'AggregateUDF', 'Alias', 'Analyze', 'Between', 'Case', 'Cast', 'Config', 'CreateMemoryTable', 'CreateView', 'DFSchema', 'DataFrame', 'Distinct', 'DropTable', 'Exists', 'Explain', 'Expr', 'Extension', 'Filter', 'GroupingSet', 'ILike', 'InList', 'InSubquery', 'IsFalse', 'IsNotFalse', 'IsNotNull', 'IsNotTrue', 'IsNotUnknown', 'IsTrue', 'IsUnknown', 'Like', 'Limit', 'Negative', 'Not', 'Partitioning', 'Placeholder', 'Projection', 'Repartition', 'RuntimeConfig', 'SQLOptions', 'ScalarSubquery', 'ScalarUDF', 'ScalarVariable', 'SessionConfig', 'SessionContext', 'SimilarTo', 'Sort', 'Subquery', 'SubqueryAlias', 'TableScan', 'TryCast', 'Window', 'WindowFrame', '__all__', '__builtins__', '__cached__', '__doc__', '__file__', '__loader__', '__name__', '__package__', '__path__', '__spec__', '__version__', '_internal', 'col', 'column', 'common', 'expr', 'lit', 'literal', 'udaf', 'udf']

>>> dir(datafusion.common)
['DFSchema', 'DataType', 'DataTypeMap', 'PythonType', 'SqlFunction', 'SqlSchema', 'SqlStatistics', 'SqlTable', 'SqlType', 'SqlView', '__builtins__', '__cached__', '__dir__', '__doc__', '__file__', '__getattr__', '__loader__', '__name__', '__package__', '__spec__']

Personally, instead of exposing just the _internal module from PyO3, I would create all these modules and submodules directly from PyO3. I'd use _datafusion as the main module, but instead of having a common.py file, the _datafusion.common submodule would be created directly from PyO3. So, all this magic of __getattr__ and __dir__ is not needed. @andygrove is there a reason why this wasn't implemented like this?

Also, I don't think it's a good practice to have everything in the main namespace. I think it's fine that for examples users have to use datafusion.expr.IsTrue instead of datafusion.IsTrue.