Implement tracing with Perfetto

[cassava]

We are currently implementing performance statistics by hand, which is fine. But when it comes to actually fine-tuning the performance and understanding what takes how long, it is useful to have a more robust method for tracing.

There is a PR #90 with more hand-implemented stuff, but it would probably be better to just use a good framework, hence Perfetto.

This partially depends on C++17, so it may make sense to wait for that to be enabled.