#40 Rewrite ArrayTimeSeries to take advantage of np storage, rewrite tests

[cocochrane]

These are the changes in ArrayTimeSeries:

1. init(): now checks sizes of times and values
2. getitem(): uses np.take() which is faster than regular indexing (says the internet)
3. setitem(): uses np.put() likewise faster than regular indexing/setting
4. len(): uses np.size()
5. values: just returns the values (doesn't need to cast into a np array again)
6. times: just returns the times (doesn't need to cast into a np array again)
7. interpolate: now uses np.searchsorted() instead of the binary_search code
8. neg(): uses numpy multiplication
9. abs(): uses numpy absolute value
10. bool(): uses np.count_nonzero()
11. add(): uses numpy addition
12. mul(): uses numpy multiplication
13. eqtimes(): uses np.array_equal()
14. eqvalues(): uses np.array_equal()

Functions omitted (should just inherit these from the parent because I can't find a faster way to implement them):

1. repr()
2. str()
3. iter()
4. contains()
5. itertimes()
6. iteritems()
7. itervalues()
8. items()
9. identity()
10. lazy()
11. pos()
12. sub()
13. eq()

@brennan-ae See anything on this bottom list that you think could be sped up? I was unsure whether I should use the numpy iterator?

[coveralls]

Coverage decreased (-0.2%) to 99.82% when pulling 454555f3d45d590549ceb22858add95667adc2c7 on ChangeArrayTS into c101a66c35b83bcee5560b0f18bee1b23ba08591 on master.