Compact historical data in different granularity #41

I like this idea!

FYI. We are implementing a slightly different strategy into a future release v0.20 right now.

• We will keep all the original data forever. So users never lose the details of the original data.

Since TT's data files are stored in folders with timestamps in names, users can easily delete old data files at any time if they decide not to keep the original data.

In the meanwhile, we provide aggregated data of min/max/sum/count (default 1 hour interval) to accelerate queries.

• If a query range is long enough, we will use aggregated data. It will be much faster.
• On the other hand, if a query range is short or the query specially wants to zoom into small granularity of data (i.e., smaller downsampling than 1 hour), we will use original data.