A two column aggregation -- Linear regression

Thanks for all the great work on datashader. It's a very impressive project.

I was wondering how hard would it be to implement a two column aggregation when drawing points. For example,
```
agg = cvs.points(df, 'lon', 'lat', ds.count('some_other_col'))
```
counts the number of points. How would I go about defining a new two column aggregation
```
agg = cvs.points(df, 'lon', 'lat', ds.linear_regression_slope('col1', 'col2'))
```
Where the slope of the linear regression between `col1` and `col2` within the bounds of the pixel was calculated instead. Obviously this would require setting a threshold where at least two (or more) observations must be present in a pixel before a slope can be defined.

I'm happy to do the work and submit a PR, I'm just wondering how involved this additional feature would be. Thanks!