Enable cross-pol power spectra *and* Python 3 support

[philbull]

This PR enables cross-polarisation support, subject to the limitation that the beam scalar can't be computed for cross-pol yet. (In other words, this PR only allows unnormalised cross-pol spectra to be computed.)

The main change in the PR is to store polarisation pairs inside UVPSpec, instead of individual polarisations. Pol-pairs can be specified as tuples, e.g. ('xx', 'xx'), or as specially-constructed integers (there are helper functions to deal with these). Individual polarisations (e.g. 'xx') are no longer accepted.

I also got a bit carried away and enabled Python 3 support in this PR. This required more extensive changes than first expected, as functions like map() and reduce() have been deprecated (now replaced with list comprehensions and np.reduce), some string handling has changed (especially annoying for HDF5), and methods like dict.keys() now return iterators instead of lists, and so can't be indexed directly (fixed by calling list() on the iterators).

[coveralls]

Coverage increased (+0.2%) to 96.654% when pulling 39d526a on enable_xpol into d2de2ff on master.

[philbull]

The tests are only failing because of a weird bug with the print statement in version.py in Python 2.x.

[philbull]

OK, I managed to fix the test failure (it was due to a change in io.StringIO behaviour between Py2 and Py3, used by the test suite). I also made some improvements in test coverage and caught a bug, so coveralls passes now too.

[nkern]

@philbull just for posterity: on the hera_pspec telecon 3/5/18 I think we decided to punt on cross-visibility-polarization scalars, and that cross-visibility-pol jacknives should be done in "data units" rather than "cosmo units"--so no scalar.

Also, I take back my support for interpreting polpairs = ['XX'] as polpairs = [('XX', 'XX')], because I think that can easily lead to confusion over 'XX' being a "polarization-pair", which it isn't in our definition, although it is technically a "feed-polarization-pair". Up to you @philbull what you want to do about it though..

[nkern]

@philbull so I've finished my skim of the PR. it looks good to me with minor requests for revision. ping me when you are done w/ edits and i'll approve!

[philbull]

Thanks Nick! I've made all the changes you suggested, and improved test coverage slightly to get that coveted green checkmark.

[nkern]

Looks great, thanks @philbull for getting us py3k supported and for the cross-visibility-pol functionality!