two bugs fixed in the calculations of pinch-zoom

[Alpt]

In dygraph-interaction-model.js, DygraphInteraction.moveTouch, the following is done:

 g.dateWindow_ = [
      c_init.dataX - swipe.dataX + (context.initialRange.x[0] - c_init.dataX) / xScale,
      c_init.dataX - swipe.dataX + (context.initialRange.x[1] - c_init.dataX) / xScale
    ];

However, it seems that swipe.dataX is not properly scaled. It should be:
c_init.dataX - swipe.dataX / xScale + (context.initialRange.x[0] - c_init.dataX) / xScale

In startTouch(), pageX, pageY is not appropriate for the calculation of dataX, dataY. This causes serious problems in the calculation of context.initialPinchCenter. The wrong calculations affect pinch-zoom when the x or y range is small and the graph div is farther to the bottom or right of the page.
For example, see http://jsfiddle.net/j63zvups/1/ and try to pinch-zoom in vertical, for scaling the y axis. You will get an ungovernable behaviour.

The solution is to use t.clientX and t.target.getBoundingClientRect().

To see that the old calculations were incorrect, consider this:

toDataYCoord(y) = yRange[0] + (area.y + area.h - y) / area.h * (yRange[1] - yRange[0]);

Now, if you pass y = pageY, you can see that (area.y + area.h - pageY) / area.h is wrong when pageY > area.y + area.h, which is the case when you scroll far enough.

With this two changes, dygraph behaves normally in my graphs and is usable.

[coveralls]

Coverage decreased (-0.02%) to 90.176% when pulling ef5fe4d on Alpt:master into e9155b2 on danvk:master.