Release the GIL in Cython code

[mrocklin]

Pandas could release the GIL in parts of the Cython code. This would enable the parallel use of multiple cores with threading.

From off-line conversation with @jreback @cpcloud

http://lbolla.info/blog/2013/12/23/python-threads-cython-gil

[TomAugspurger]

FYI: scinkit-image just did this in scikit-image/scikit-image#1519.

[mrocklin]

It happened surprisingly quickly in their case. I think that it was mostly the result of an afternoon sprint of a couple developers followed by a couple days of cleanup.

[jreback]

@cpcloud did some prelim work on this. I don't think its too difficult. Just requires some effort.

[shoyer]

Indeed, it mostly just takes a little bit of testing work. Glad to see this being done!

For what it's worth, I looked into releasing the GIL a little bit in the hashtable/index API as part of my IntervalIndex work (#8707). The main stumbling block I encountered was pandas.hashtable.Int64Vector (and the other Vector objects). Hopefully someone more clever than me could figure out how to release the GIL on the append method by moving the resize call to use the C API: https://github.com/pydata/pandas/blob/0aceb38ac1a4659ae298d726bc447b97943163dc/pandas/hashtable.pyx#L68