Column access to tables

[hyanwong]

With #17 we have defined a simple table structure, but we have no simple way to get a numpy array of the columns, for instance, we might hope to be able to get

child_left = TableCollection.intervals.child_left

Or even set a whole column:

TableCollection.intervals.child_left = child_left +1

(we can't even do the latter in tskit, though)

[duncanMR]

I'd be happy to take this one on. I will see if I can replicate how its done with tskit data.

[hyanwong]

Great. We could probably do a bit better than tskit for the interface. For example, it would be good to have something like this:

tskit-dev/tskit#2015

It would be much better if we could use tables.nodes.time[x] to get direct memory access, rather than the somewhat clunky ts.nodes_time[x]. You might well want to talk to @benjeffery about this.

Do make sure, however, that we aren' "prematurely optimising". For the size of datasets we are experimenting on, there's no need to make this fast (yet). It's better to get a slow API that can be sped up later, under the hood.

Once we get to this stage, it's vital to start setting up unit tests, so #20 is important to tackle at the same time.