use more accurate measure of average for frametime

[tybug]

we use the median frametime currently, which can be misleading, for instance on a replay like this -

median could end up being right on the edge of one of the spikes and that information does not carry over into the value.

A better approach would be to split the (sorted) frametimes into 4 blocks, take the median of each, and then take the mean of the four medians.