Week 7

The existing PR is passing Travis tests but hasn't been reviewed yet, so in the meantime I've been working on some performance issues. One huge problem I discovered last week was that `group_by` created a number of Table slices, each of which incurred an expensive deep copy of column indices and contributed to a running time of several seconds. To circumvent the problem, I created a context manager `static_indices`, which can be used like so:
```
with static_indices(table):
    subtable = table[[1, 3, 4]]
```
In this case `subtable` will have no indices. However, the main issue was with column slicing, which should represent a reference rather than a (deep) copy in keeping with the behavior of `numpy.ndarray` (and thus `Column` internal data). Lucky that this came up, because I didn't have any previous tests checking to make sure that modifying a slice affects both the internal data and indices of the original Table or Column.

Aside from this, I've been working on cutting down the time required to initialize an index; the pure-Python loop I previously used was woefully inefficient memory- and time-wise. I haven't yet figured out how to get this to work with FastRBT, but SortedArray is now much faster to initialize. Here are some time profiling results on a 100,000 line table:
```
FastRBT
**********
time_init: 1.2263660431
time_group: 0.2325990200
time_where: 0.0003449917
time_query: 0.0000329018
time_query_range: 0.0643420219
time_add_row: 2.7397549152
time_modify: 0.0001499653

SortedArray
**********
time_init: 0.0355048180
time_group: 0.0041830540
time_where: 0.0000801086
time_query: 0.0000169277
time_query_range: 0.0217781067
time_add_row: 0.0200960636
time_modify: 0.0016808510

None
**********
time_init: 0.0000019073
time_group: 0.0865180492
time_where: 0.0002820492
time_query: 0.0001931190
time_query_range: 0.2128648758
time_add_row: 0.0006089211
time_modify: 0.0000159740

```

I've focused on SortedArray quite a bit, so FastBRT is still pretty slow in areas that should be easily fixable--I'll tackle those tomorrow. I have an IPython notebook with these timing results here, and a short Markdown file here.