Benchmark mode

It'd be cool if there was a mode where magic-trace would take a snapshot for some configurable time span, and then you use can use that snapshot to iteratively display histograms and summary statistics of different functions' runtimes.

Perfetto supports SQL queries on generated traces and I initially thought it'd be best to use that to build something fully-featured here. It's certainly convenient, but I think the user experience would be better if they had more immediate feedback in their terminals instead of having to boot up the GUI.

The issue names no files, tests, or entry points. Start by mapping how magic-trace generates traces and how Perfetto SQL queries and the GUI currently expose them; then define the terminal workflow, configurable capture span, histogram and summary outputs, and tests that demonstrate the completed mode.

Written by the indexing model from the issue text.