the flamegraph view can consume a ton of memory when opening a large data file with a big call graph tree. we currently build all items for all frames in the stack, but only display those that are costly enough for the current zoom level. instead, we should build the items only as-needed, and reuse them also when zooming to prevent costly reallocation of the items.

note: the flame graph items are now cut-off below a user-configurable threshold, which gives good performance. Still, the above would be even nicer to have.