When hovering over a time series user chart the performance can be poor and cpu usage spikes.
This is particularly evident when the number of points are high for activities with 1s samples > 2-3hrs.

The scatter plot uses a quadtree to assist in mapping the mouse cursor to data points and is highly performant. But the time series not only uses lower_bound to search the entire series (and repeats this for EVERY series) it also takes a copy of data points to achieve it. (not sure why).

The code needs to be rewritten to consider performance since time series charts are the single most popular way of visualising activity data.

See code at GenericSelectTool:681-723