arrow

ARROW-3890: [Python] Handle NumPy binary arrays with UTF-8 validation when converting to StringArray

I'm not sure if all compilers will be smart enough to do loop unswitching here. If it ends up being a bottleneck I suggest rewriting in a follow up patch.

The BinaryArray overflow issue (ChunkedArray is not being produced) is still present here. We will need to address that in ARROW-2970

This patch also includes symbol export macros particular to the arrow_python shared library. These are needed so that global data members in arrow.dll can be accessed in arrow_python.dll

Author: Wes McKinney <wesm+git@apache.org>

Closes #3063 from wesm/ARROW-3890 and squashes the following commits:

dac4995f9 <Wes McKinney> Windows needs arrow.lib in addition to arrow_python.lib now because of the new export flags
91dbea8bb <Wes McKinney> Add libarrow_python-specific visibility macros so that global data members from arrow.dll can be accessed correctly in arrow_python.dll
062c3836f <Wes McKinney> Clarify comment
cfbd30bbe <Wes McKinney> Handle case where user passes UTF-8 encoded numpy.str_ type array to pyarrow.array with type=pyarrow.string()