Add some minimal trend analysis to cantal

[tailhook]

To make scheduling decisions we need two things: absolute value of some metric and a trend.

Currently, we fetch all the datapoints and compute math on top. But it's better to aggregate values in cantal itself. We already have aggregation function facility. We just need a few functions:

1. Gradient value over N data points fit using least squares
2. Moving average over N data points
3. Maximum value over N data points (probably applied after moving average)