Numpy support about turbodbc

[MathMagique]

Released with turbodbc 0.4.0

[MathMagique]

This requires a major rework of the result set class and others. Key idea for the new design:

• query allows to get current result_set
• A result_set is rather basic. It creates the required buffers, binds them to the columns, and implements batched access with something like next_batch(). Also provides access to the raw multi_value_buffers
• A row_based_result_set is based on a result_set and takes care of handling the row concept. Grants access to raw memory in terms of buffer_elements
• A cpp_row_based_result_set is based on a row_based_result_set and adds conversion from SQL data type to turbodbc::field.

Above components merely shift the responsibilities around without creating new user value. Below components can be added to implement new features:

• A py_row_based_result_set is based on row_based_result_set and replaces the SQL data type -> turbodbc::field -> Python type conversion with a direct hop from SQL data type -> Python type. Should improve performance a little.
• A numpy_based_result_set is based on the plain result_set and contains specialized routines for copying entire chunks of memory without creating individual C++ or Python objects.

[MathMagique]

Found a way to make numpy support both available and purely optional, so folks can still use turbodbc without having to install and compile numpy first.