Add interpolator field to TimeSeries

[tuckermcclure]

We've talked about this, and here it is. It's very simple, and that's the goal. It also has a useful default. I tested that this powers our downstream packages with no changes.

[CarlOlsson]

Two questions

1. If we specify an non default interpolator on a time series ts1, save it to hdf5 and then load it again as ts2, will the loaded time series still evaluate to the same, ie. is ts1(0.5) == ts2(0.5)

2. I am wondering about the custom interpolator with states. Say an interpolator returns linear interpolation plus 10. Now ts(1:0.1:10) will retrun ts_linear_interp(1:0.1:10) .+ 10. That is good but what happens for ts2 = ts(1:0.1:10), ts3 = ts2(3.0:0.1:4.0). This will again add 10?

[tuckermcclure]

Right now, if you save the logs to HDF5 and then load them, the interpolator information is lost. I think we should add that. I did not opt to add that in this PR, but I would do that soon. The way I would do it is that I would serialize the whole interpolator and deserialize it on load. This is not flexible. To use the log in the future, you'd have to have the same interpolator defined. That's already true for a bunch of the types that are serialized though, so I'm not too worried about the portability of the interpolator. Would you want that as a feature of this PR, or is a follow-up PR a better place for that?

[CarlOlsson]

All good, just wanted to make sure we didn't miss anything.