Chrome OS has a feature for detecting browser hangs.

session_manager periodically sends a D-Bus message to the browser process, and it must respond within a certain time, or the browser is killed (via SIGABRT).

We do get crash reports from chrome for the abort, but it's not obvious that they are triggered by a browser hang. Additionally, the instruction pointers could vary wildly for the same cause, so it's not possible to aggregate the crashes back for any hang statistics.

Kirtika suggested that we report browser hangs to crash/ much like kernel warnings, which seems like a good approach.

This has an additional benefit over reporting statistics in UMA in that we should be able to match up the Chrome-reported crash pretty easily by timestamp.

Any objections?