[Python] Multithreaded conversion between Arrow record batches as NumPy arrays (for pandas)

[asfimport]

This only impacts columns that are not zero-copyable. The GIL only needs to be held while invoking C API functions like PyArray_SimpleNew or PyString_FromStringAndSize. If there are many string columns, we may be concerned over GIL contention

Reporter: Wes McKinney / @wesm

_{Note: This issue was originally created as ARROW-364. Please see the migration documentation for further details.}

[asfimport]

Wes McKinney / @wesm:
Duplicate of ARROW-428