API/ENH: add support for integer NA with dynd-python dep

[jreback]

this should actually be relatively straightforward.

Incorporating this library: https://github.com/libdydn/dynd-python
as an optional dep, allows native Integer NA support (implemented
very similarly to how NaT is represented in datetime64[ns] dtypes ATM.

dynd-python are the python bindings for libdynd and have an almost 100% numpy
compatibility (they use the buffer interface so should be perf-wise pretty costless to do this).

tangentially related to #8350

cc @mwiebe
cc @shoyer
cc @jorisvandenbossche
cc @cpcloud

[mwiebe]

To comment on the #8350 tangential relation, one way to think of libdynd is as an experimental sandbox in which array programming ideas similar to numpy but without a strong backwards compatibility guarantee are being evolved. This means, proportional to development resources available, libdynd can quickly react to pandas' needs and implement changes. The flipside of this coin is that any library depending on libdynd also needs to keep track of changes there as it evolves to the various demands from different use cases and projects.

What we have been doing with blaze is keeping the head of master of each working together, validating that they work together via travis-ci, and doing releases at roughly the same time.

[datnamer]

Is this still planned?

[jreback]

at some point

[jorisvandenbossche]

@jreback Let's close this issue in favor of discussions in the pandas2 repo?